Mathematics

Advances in Applied Mathematics

FOR
DYNAMICAL SYSTEMS
BIOLOGICAL MODELING
DYNAMICAL SYSTEMS
FOR BIOLOGICAL
MODELING
Dynamical Systems for Biological Modeling: An Introduction pro-
vides both biology and mathematics students with the understanding
and techniques necessary to undertake basic modeling of biological
systems. It achieves this through the development and analysis of
dynamical systems. AN INTRODUCTION
The approach emphasizes qualitative ideas rather than explicit com-
putations. Some technical details are necessary, but a qualitative ap-
proach emphasizing ideas is essential for understanding. The model-
ing approach helps students focus on essentials rather than extensive
mathematical details, which is helpful for students whose primary in-
terests are in sciences other than mathematics.
The book discusses a variety of biological modeling topics, including
population biology, epidemiology, immunology, intraspecies competi-
tion, harvesting, predator–prey systems, structured populations, and
more.
The authors also include examples of problems with solutions and
some exercises that follow the examples quite closely. In addition,
problems are included that go beyond the examples, both in math- Fred Brauer
Brauer • Kribs
ematical analysis and in the development of mathematical models for
biological problems, in order to encourage deeper understanding and
an eagerness to use mathematics in learning about biology. Christopher Kribs

C664X

w w w. c rc p r e s s . c o m

C664X_cover.indd 1 11/9/15 3:23 PM

DYNAMICAL SYSTEMS
FOR BIOLOGICAL
MODELING
AN INTRODUCTION
Advances in Applied Mathematics

Series Editor: Daniel Zwillinger

Published Titles
Green’s Functions with Applications, Second Edition Dean G. Duffy
Introduction to Financial Mathematics Kevin J. Hastings
Linear and Integer Optimization: Theory and Practice, Third Edition
Gerard Sierksma and Yori Zwols
Markov Processes James R. Kirkwood
Pocket Book of Integrals and Mathematical Formulas, 5th Edition
Ronald J. Tallarida
Stochastic Partial Differential Equations, Second Edition Pao-Liu Chow
Dynamical Systems for Biological Modeling: An Introduction
Fred Brauer and Christopher Kribs
 Advances in Applied Mathematics

DYNAMICAL SYSTEMS
FOR BIOLOGICAL
MODELING
AN INTRODUCTION

Fred Brauer
University of British Columbia
Vancouver, British Columbia, Canada

Christopher Kribs
University of Texas at Arlington
Arlington, Texas, USA
Back cover image credit: (Photo of healthcare workers) Cleopatra Adedeji, CDC PHIL public domain
collection.

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2016 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20151221

International Standard Book Number-13: 978-1-4987-7404-8 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copy-
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro-
vides licenses and registration for a variety of users. For organizations that have been granted a photo-
copy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
to our children and grandchildren
This page intentionally left blank
Contents

Preface xi

Acknowledgments xiii

I Elementary Topics 1
1 Introduction to Biological Modeling 3
1.1 The nature and purposes of biological modeling . . . . . . . 3
1.2 The modeling process . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Types of mathematical models . . . . . . . . . . . . . . . . . 11
1.4 Assumptions, simplifications, and compromises . . . . . . . . 14
1.5 Scale, and choosing units . . . . . . . . . . . . . . . . . . . . 17

2 Difference Equations (Discrete Dynamical Systems) 23

2.1 Introduction to discrete dynamical systems . . . . . . . . . . 23
2.1.1 Linear difference equations . . . . . . . . . . . . . . . 24
2.1.2 Solution of linear difference equations . . . . . . . . . 26
2.1.3 Nonlinear difference equations . . . . . . . . . . . . . . 28
2.2 Graphical analysis . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Qualitative analysis and population genetics . . . . . . . . . 37
2.3.1 Linearization and local stability . . . . . . . . . . . . . 37
2.3.2 A problem in population genetics . . . . . . . . . . . . 42
2.4 Intraspecies competition . . . . . . . . . . . . . . . . . . . . 50
2.4.1 Two metered fish models . . . . . . . . . . . . . . . . 54
2.4.2 Between contest and scramble . . . . . . . . . . . . . . 56
2.5 Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.1 Fishery harvesting and graphical equilibrium analysis 60
2.6 Period doubling and chaos . . . . . . . . . . . . . . . . . . . 69
2.6.1 Dispersal . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.6.2 Dynamical diseases and physiological control systems 78
2.7 Structured populations . . . . . . . . . . . . . . . . . . . . . 82
2.7.1 Spatial dispersal . . . . . . . . . . . . . . . . . . . . . 86
2.7.2 Two-stage populations . . . . . . . . . . . . . . . . . . 90
2.8 Predator-prey systems . . . . . . . . . . . . . . . . . . . . . . 93
2.8.1 A plant-herbivore model . . . . . . . . . . . . . . . . . 93
2.8.2 A host-parasitoid model . . . . . . . . . . . . . . . . . 94
Miscellaneous exercises . . . . . . . . . . . . . . . . . . . . . . . . 97

vii
viii Contents

3 First-Order Differential Equations (Continuous Dynamical

Systems) 99
3.1 Continuous-time models and exponential growth . . . . . . . 99
3.1.1 Exponential growth . . . . . . . . . . . . . . . . . . . 100
3.1.2 Radioactive decay . . . . . . . . . . . . . . . . . . . . 105
3.2 Logistic population models . . . . . . . . . . . . . . . . . . . 108
3.2.1 All creatures great and small? . . . . . . . . . . . . . . 110
3.2.2 Competition among plants . . . . . . . . . . . . . . . 114
3.2.3 The spread of infectious diseases . . . . . . . . . . . . 117
3.3 Graphical analysis . . . . . . . . . . . . . . . . . . . . . . . . 123
3.3.1 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.3.2 Direction fields . . . . . . . . . . . . . . . . . . . . . . 126
3.4 Equations and models with variables separable . . . . . . . . 130
3.4.1 A linear model for the cardiac pacemaker . . . . . . . 131
3.4.2 General procedure . . . . . . . . . . . . . . . . . . . . 135
3.4.3 Solution of logistic equations . . . . . . . . . . . . . . 139
3.4.4 Discrete-time metered population models . . . . . . . 141
3.4.5 Allometry . . . . . . . . . . . . . . . . . . . . . . . . . 144
3.5 Mixing processes and linear models . . . . . . . . . . . . . . 150
3.5.1 Chemostats . . . . . . . . . . . . . . . . . . . . . . . . 152
3.5.2 Drug dosage . . . . . . . . . . . . . . . . . . . . . . . . 155
3.5.3 Newton’s law of cooling . . . . . . . . . . . . . . . . . 158
3.5.4 Migration . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.6 First-order models with time dependence . . . . . . . . . . . 165
3.6.1 Superposition . . . . . . . . . . . . . . . . . . . . . . . 165
3.6.2 Integrating factors . . . . . . . . . . . . . . . . . . . . 167
3.6.3 Substitution and integration . . . . . . . . . . . . . . . 169
3.6.4 Mixing processes with variable coefficients . . . . . . . 171
3.6.5 Bernouilli equation . . . . . . . . . . . . . . . . . . . . 174
Miscellaneous exercises . . . . . . . . . . . . . . . . . . . . . . . . 182

4 Nonlinear Differential Equations 185

4.1 Qualitative analysis tools . . . . . . . . . . . . . . . . . . . . 185
4.1.1 Possible end behaviors . . . . . . . . . . . . . . . . . . 187
4.1.2 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.1.3 Phase portraits . . . . . . . . . . . . . . . . . . . . . . 196
4.1.4 Foraging ants and phase transitions . . . . . . . . . . 199
4.2 Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4.2.1 Constant-yield harvesting . . . . . . . . . . . . . . . . 203
4.2.2 Constant-effort harvesting . . . . . . . . . . . . . . . . 214
4.2.3 Migration and dispersal as harvesting . . . . . . . . . 222
4.2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 224
4.3 Mass-action models . . . . . . . . . . . . . . . . . . . . . . . 228
4.3.1 A simple chemical reaction . . . . . . . . . . . . . . . 228
4.3.2 The spread of infectious diseases, revisited . . . . . . . 231
 Contents ix

4.3.3 Contact rate saturation and the “Pay It Forward” model 237
4.4 Parameter changes, thresholds, and bifurcations . . . . . . . 242
4.4.1 Hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . 249
4.4.2 The spruce budworm . . . . . . . . . . . . . . . . . . . 251
4.5 Numerical analysis of differential equations . . . . . . . . . . 258
4.5.1 Approximation error . . . . . . . . . . . . . . . . . . . 259
4.5.2 Euler’s method . . . . . . . . . . . . . . . . . . . . . . 260
4.5.3 Other numerical methods . . . . . . . . . . . . . . . . 264
4.5.4 Eutrophication . . . . . . . . . . . . . . . . . . . . . . 270
Miscellaneous exercises . . . . . . . . . . . . . . . . . . . . . . . . 277

II More Advanced Topics 281

5 Systems of Differential Equations 283
5.1 Graphical analysis: The phase plane . . . . . . . . . . . . . . 283
5.2 Linearization of a system at an equilibrium . . . . . . . . . . 291
5.3 Linear systems with constant coefficients . . . . . . . . . . . 297
5.3.1 A liver chemistry example . . . . . . . . . . . . . . . . 305
5.4 Qualitative analysis of systems . . . . . . . . . . . . . . . . . 311
Miscellaneous exercises . . . . . . . . . . . . . . . . . . . . . . . . 319

6 Topics in Modeling Systems of Populations 321

6.1 Epidemiology: Compartmental models . . . . . . . . . . . . . 321
6.1.1 An epidemic model . . . . . . . . . . . . . . . . . . . . 321
6.1.2 A model for endemic situations . . . . . . . . . . . . . 327
6.2 Population biology: Interacting species . . . . . . . . . . . . 331
6.2.1 Species in competition . . . . . . . . . . . . . . . . . . 331
6.2.2 Predator-prey systems . . . . . . . . . . . . . . . . . . 337
6.2.3 Symbiosis . . . . . . . . . . . . . . . . . . . . . . . . . 345
6.3 Numerical approximation to solutions of systems . . . . . . . 351
6.3.1 Example: A two-sex model . . . . . . . . . . . . . . . 352

7 Systems with Sustained Oscillations and Singularities 359

7.1 Oscillations in neural activity . . . . . . . . . . . . . . . . . . 359
7.1.1 The Fitzhugh-Nagumo equations . . . . . . . . . . . . 360
7.1.2 A model for cat neurons . . . . . . . . . . . . . . . . . 363
7.2 Singular perturbations and enzyme kinetics . . . . . . . . . . 366
7.2.1 Bursting . . . . . . . . . . . . . . . . . . . . . . . . . . 372
7.2.2 An example from enzyme kinetics . . . . . . . . . . . 374
7.3 HIV: An example from immunology . . . . . . . . . . . . . . 379
7.3.1 A basic model . . . . . . . . . . . . . . . . . . . . . . 381
7.3.2 Including infected cells . . . . . . . . . . . . . . . . . . 387
7.4 Slow selection in population genetics . . . . . . . . . . . . . . 392
7.4.1 Equally fit genotypes . . . . . . . . . . . . . . . . . . . 393
7.4.2 Slow genetic selection . . . . . . . . . . . . . . . . . . 396
x Contents

7.5 Second-order differential equations: Acceleration . . . . . . . 402

7.5.1 The harmonic oscillator . . . . . . . . . . . . . . . . . 402
7.5.2 The van der Pol oscillator . . . . . . . . . . . . . . . . 405
7.5.3 A model of oxygen diffusion in muscle fibers . . . . . . 408

III Appendices 413

A An Introduction to the Use of MapleTM 415
A.1 Plotting graphs of functions . . . . . . . . . . . . . . . . . . 416
A.2 Graphical solution of first-order differential equations . . . . 417
A.3 Graphical solution of systems of differential equations . . . . 419
A.4 The cobwebbing method for graphical solution of first-order
difference equations . . . . . . . . . . . . . . . . . . . . . . . 420
A.5 Solution of difference equations and systems of difference equa-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
A.6 A bifurcation program . . . . . . . . . . . . . . . . . . . . . . 424

B Taylor’s Theorem and Linearization 425

C Location of Roots of Polynomial Equations 427

D Stability of Equilibrium of Difference Equations 429

Answers to Selected Exercises 433

Bibliography 459

Index 467
Preface

Some understanding of difference equations and differential equations is be-

coming essential for students in the biological sciences. There are several recent
texts providing an introduction to mathematical biology through dynamical
systems, but most of these are accessible primarily to students of mathematics
with some interest in applications to biology rather than to biology students
who have not (yet) developed a disposition to describe real systems in mathe-
matical terms. Many of these texts present fully developed models of biological
systems ready for mathematical analysis. Such models provide students with
exercises in applying mathematical techniques but little experience in the
(often iterative) translation of biological concepts into mathematical terms
and vice versa, which is at the heart of modeling. It is our intention to try
to address both of these issues, to prepare students of biology as well as of
mathematics with the understanding and techniques necessary to undertake
basic modeling of biological systems through the development and analysis of
dynamical systems.
We propose to present an introduction to dynamical systems, covering both
discrete (difference equation) and continuous (differential equation) types, for
students who have had an introduction to calculus. While we assume that
students have learned basic material including differentiation, integration, and
exponential and logarithmic functions, we will recall topics which may not
have made sufficient impression to have been absorbed completely and will
describe some topics from a slightly different perspective. Our motivation will
come from biological topics, including but not confined to population biology
and epidemiology. Our approach will emphasize qualitative ideas rather than
explicit computations; we feel that this approach is both easier for students
whose primary interests are not in mathematics and more useful in many
applications. This material is not, however, intended to serve as a differential
equations course for students in the physical sciences as it does omit many
techniques and topics that are essential for such students.
In presenting mathematical topics, we will attempt to tell the truth and
nothing but the truth, but not necessarily the whole truth. We will try to
emphasize the basic truths but not to overwhelm the student with precise
technical detail. Some results will be stated without proof, while others will
be accompanied by outlines of the reasons why they are true. We will normally
avoid detailed, rigorous proofs.
Mathematics is not a spectator sport, and can be learned only by solving

xi
xii Preface

problems. We include examples of problems with solutions and some exercises

which follow the examples quite closely. However, a library of solved examples
used as templates will not be sufficient to meet the needs of a developing sci-
entist. For this reason, we also include problems that go beyond the examples,
both in mathematical analysis and in the development of mathematical mod-
els for biological problems, in order to encourage deeper understanding and
an eagerness to use mathematics in learning about biology. We also include
some problems, marked with an asterisk (*), which are more challenging.
We recommend the introduction to modeling in Chapter 1 as a beginning
of any course on biological modeling using this text. The contents of Chapters
2 through 4 overlap considerably with the material which would be covered in
a unified course (probably two semesters in length) which covers calculus and
some difference equations and elementary differential equations. For students
who have already taken such a course, a suitable course could review these
topics as needed (after starting with Chapter 1) and then continue with Chap-
ters 5, 6, and possibly 7. For students who have had a semester of calculus
without difference or differential equations, a suitable course could consist of
the first four chapters.
We provide some support for helping readers use software packages to
generate numerical solutions and graphs (as we ourselves have done to make
some of the figures in this book), primarily in the sections dealing with nu-
merical analysis and in the appendices. These packages and the companies
that produce them are listed below.

Maple is a registered trademark of Waterloo Maple, Inc., www.maplesoft.com.

Mathematica is a registered trademark of Wolfram Research, Inc., at

www.wolfram.com.

MATLAB is a registered trademark of The Mathworks, Inc. For prod-

uct information please contact The Mathworks, Inc., 3 Apple Hill Drive,
Natick, MA 01760-2098 USA, tel. 508-647-7000, fax 508-647-7001, e-mail
[email protected], www.mathworks.com.

Trademarked names may be used in this book without the inclusion of a

trademark symbol. These names are used in an editorial context only; no
infringement of trademark is intended.
Acknowledgments

We thank Bob Stern at CRC Press for planting the initial idea for this book,
and both him and Bob Ross for their support and understanding in the face
of long delays during the preparation of the manuscript. (Although we do not
include delay equations in this book, there certainly were delays.) We also
thank everyone at Taylor & Francis who helped with the production of this
book, especially Shashi Kumar and Marcus Fontaine for critically useful help
involving LaTeX, Karen Simon, and Kevin Craig for graphic design assistance.
The figures and photos accompanying discussion of the various biologi-
cal systems studied and discussed play a crucial role in bringing the models
(and their motivations) to life for the reader, and many appear in this work
through the kind permission of others. We therefore acknowledge here all those
who generously permitted us to print their photos, or helped us obtain permis-
sion: Donna Anstey, Francine Bérubé, Daniel Bowen, John Calambokidis, Tom
Chrzanowski, Patricia Ernst, Carla Flores, Tim Gerrodette, Stefano Guerrieri,
Ray Hamblett, Alan M. Hughes, Duncan Jackson, Russell S. Karow, Carolyn
Kribs, Joel Michaelsen, Bernard E. Picton, Dave Powell, Francis Ratnieks,
Helen Sarakinos, Howard Swatland, Michael Tildesley, and S. Bradleigh Vin-
son.
We also thank all those photographers who contributed indirectly, in-
cluding photos in the public domain: Cleopatra Adedeji, Lennert B., David
Burdick, Janice Haney Carr, Mark Conlin, Karen Couch, Jan Derk, Gary
Fellers, Ryan Hagerty, William Chapman Hewitson, Steve Hillebrand, Tom
Hodge, John & Karen Hollingsworth, Al Mare, Maureen Metcalfe, Benjamin
Mills, A.J. Nicholson, Sergey Nivens, Zev Ross, Jeff Schmaltz, Greg Webster,
Gary Zahm, and the following organizations: the Centers for Disease Con-
trol, Florida Keys National Marine Sanctuary, Ken Gray Image Collection
at Oregon State University, MODIS/NASA, National Diabetes Information
Clearinghouse, NOAA, the Pennsylvania Dept. of Conservation and Natural
Resources, River Alliance of Wisconsin, the Sickle Cell Foundation of Georgia,
the Southwest Fisheries Science Center of the NOAA Fisheries Service, U.S.
Fish and Wildlife Service, U.S. Geological Survey, U.S. National Park Service,
and the Wisconsin Dept. of Natural Resources.

xiii
This page intentionally left blank
 Part I

Elementary Topics

1
This page intentionally left blank
Chapter 1
Introduction to Biological Modeling

1.1 The nature and purposes of biological modeling

Mathematical modeling has been used to explain biological systems for
centuries, going back to work such as that of Thomas Malthus (1798) on the
growth of populations, Daniel Bernouilli (1760) on smallpox vaccination and
arguably Fibonacci (Leonardo of Pisa, 1202) on the well-known rabbit prob-
lem. Modern mathematical biology has been an ongoing scientific endeavor
since the end of the nineteenth century, growing out of the work of individuals
such as P.D. En’ko in epidemiology, D’Arcy Wentworth Thompson in change
in biological organisms, and G. Stokes in fluid flows, and the technical develop-
ments of the late twentieth century have placed it in the forefront of scientific
research. The use of theoretical models to describe and predict the behavior of
biological systems is especially useful today in examining large-scale questions
where controlled experiments are either impossible or unethical to carry out
— the effects of large-scale alterations in environmental conditions on local
ecology, or of slight decreases in efficiency of certain disease control measures.
Even in situations where some experimental data is available, mathematical
models may validate existing theories or provide new insights by describing
the mechanisms through which biological processes occur, taking in a whole
spectrum of possible measurements in a single step. More specifically, one
might use modeling
• to predict unknown behavior (such as intervention effects) based on
given factors,
• to generate virtual experiments that inform general biological theory,
• to account for observed behavior as simply as possible,
• to compare competing hypotheses — which better accounts for observed
results?
• to evaluate or rule out hypotheses,
• to suggest conjectures and new hypotheses, or

3
4 Dynamical Systems for Biological Modeling: An Introduction

• to guide empirical inquiry — does this process make a key difference in

the behavior of the system?1
In short, modeling helps us consider the question, “What if ...?” The language
of mathematics, meanwhile, provides powerfully compact descriptions of com-
plicated ideas, and is capable, at its best, of offering equally simple insights,
such as the idea of a single threshold quantity which determines the outcome
of a complex biological process and brings together all the relevant factors in
a way that clarifies the relative importance of each (cf. Section 4.4).
Mathematical models, however, are far from perfect: as intricate as they
may appear at times, they are mere sketches or caricatures of the living world.
Biological systems are complex, with many underlying forces and interactions,
heterogeneity, and random variations and fluctuations. In modeling such a
system, we can only hope to develop a good enough approximation to reality
to accomplish our purposes. The engineering statistician George Box wrote,
“All models are wrong, but some are useful.”2 We use models to give rough
descriptions of reality in cases where rough is still accurate enough to give
useful insights.
By its nature, modeling involves both iteration and compromise. As model-
ing is imperfect, the results of a given model may not agree with observations,
or may have limitations that keep the model from being useful. In these cases,
the modeler must make changes to the model, or even replace it altogether,
and then analyze the new model to see if it describes the biology better (or
well enough). One has to be willing to tinker and try again. Iteration thereby
makes the modeling process cyclical rather than linear. The following section
sketches this iterative modeling process.
Section 1.3 goes into more detail about the types of mathematical models
one might use. The selection of a particular kind of mathematical structure as
a model follows from one’s research questions. A laboratory or field biologist,
interested in the specific biology of the organisms in the system, may be most
interested in quantitative or numerical results, and models which fit observed
data closely. Theoretical biologists, on the other hand, may be more interested
in qualitative, structural insights, which can suggest trends across models of a
number of similar systems, such as the role of genetic diversity in the survival
of populations. In each case, the researcher’s perspective and interests will
shape the research question and, through it, the type(s) of model and analysis
most likely to provide fruitful answers. The types of models we shall explore
in this text are called dynamical systems, which mean that they change as
functions of some independent variable(s), usually time.
Facing up to the imperfection of any model means acknowledging its lim-
itations. Modeling does not mean ignoring or brushing aside the assumptions
1 cf. E. Smith, S. Haarer, J. Confrey (1997). Seeking diversity in mathematics educa-

tion: mathematical modeling in the practice of biologists and mathematicians, Science and
Education 6(5): 441–472.
2 G.E.P. Box (1976). Science and statistics, Journal of the American Statistical Associ-

ation 71: 791–799.

 Introduction to Biological Modeling 5

and simplifications we make in constructing and interpreting models; rather,

it is an active exercise in compromise: including as much detail as necessary
while keeping the model manageable in terms of analysis and (as applica-
ble) data collection. A saying attributed to Albert Einstein runs, “Everything
should be made as simple as possible, but not simpler.” Section 1.4 discusses
some of the most common compromises faced in mathematical modeling, as
well as their consequences.
Finally, the modeling process should also include an explicit choice of scale
and unit(s) of measurement for the quantities or populations being modeled,
and Section 1.5 introduces two related issues which are discussed further as
they recur later in the text. In each section, we shall speak of biological systems
in general, as the focus here is on modeling, but the reader can substitute
specific systems, such as the human circulatory system, the current generation
of Monarch butterflies, or the population of Toronto at risk for SARS in March
2003, for the word “system” where it occurs.

1.2 The modeling process

Although mathematical modeling may look like a great variety of different
things, there are certain commonalities in the process, in the “big picture,”
and as this text is aimed at introducing students to biological modeling, it is
just as important to become familiar with the higher-order tools and elements
of modeling as with the mathematical techniques specific to each type of
model. As mentioned in the previous section, modeling is an iterative process.
Figure 1.1 sketches the flow of this process.
In practice, we begin by developing our first try at a model (which process
includes defining the problem at hand), translating the biological system into
a mathematical system. We then apply mathematical tools to analyze the
model. The mathematical results must then be interpreted back into biological
terms, and finally we must evaluate the answers our model has given us, to
decide whether they are satisfactory — biologically reasonable, and sufficient
to meet our objectives. If they are not (which is the case often enough), we
must adjust the model to try to remedy the observed shortcomings, and repeat
the process until the results are satisfactory.

modeling analysis interpretation

BIOLOGICAL MATHEMATICAL MATHEMATICAL BIOLOGICAL
PROBLEM PROBLEM RESULTS RESULTS

evaluation

FIGURE 1.1: Mathematical modeling as an iterative process.

6 Dynamical Systems for Biological Modeling: An Introduction

Mathematical biologists therefore need to be able to:

1. take biological features and hypotheses and turn them into mathematical
structures;
2. apply relevant mathematical analysis tools;
3. interpret mathematical results in biological terms; and
4. evaluate biological results critically in order to evaluate the underlying
mathematical model.

Let us consider each of these activities one by one.

(1) Modeling What does model construction look like? The details can
vary widely, but the philosophy is to describe the essential characteristics
of your system in mathematical terms. For us, this will typically mean a
population or group of interacting populations, although we will apply the
term population to collections of anything from molecules to households. One
perspective on modeling is that there are two directions in which to go about
it:
• forward — begin with observed characteristics, factors, rates, make your
model reflect these, and see if the results match observations; or
• backward — use observations to develop known (desired) model behav-
ior, and craft a model known to exhibit this behavior.
In other words, one might begin by assembling model elements which cor-
respond to the biological features of the system and consider the results (a
predictive model), or one might instead begin with a model of a type known
to behave as desired, and modify the structure to fine-tune (or fit data) (a
descriptive model). If we consider the three purposes suggested earlier for bi-
ological modeling, we see that a forward modeling approach suits the goal
of predicting unknown behavior based on known forces, while a backward (or
back-end) approach might best suit the goal of accounting for observed behav-
ior. Evaluating competing hypotheses may require a combination of the two
approaches. It is important, however, to acknowledge the distinction between
first principles models derived directly from known laws (common in physics
and chemistry, e.g., using Newton’s Laws, but less so in biology) and ad hoc
models which use terms, expressions or functions commonly used to represent
a particular feature or process of the system based on past observations of
their fit to data (e.g., using a logistic model to describe population growth
with limited resources (see Sections 2.4 and 3.2), or a mass-action term to
describe the interaction of two populations (see Section 4.3). Likewise there is
a distinction between broadly descriptive models, which attempt to capture as
much detail as possible in a biological system, and focused theoretical models
 Introduction to Biological Modeling 7

which incorporate only those features relevant to a specific research question.3

Of course, this perspective for developing models is by no means exclusive,
and should be considered only as long as it is helpful in providing direction.
The elements of model development include identifying the problem or
research question as specifically as possible (a complex task we shall not do
more than illustrate), writing careful definitions of relevant terms (variables
and parameters), specifying all the assumptions being made (see Section 1.4
for a fuller discussion), drawing diagrams (sometimes), choosing an appropri-
ate mathematical structure (see Section 1.3) and scale (see Section 1.5), and
assembling all these elements into a coherent mathematical description (for
us, an equation or set of equations). It is also important to articulate one’s
objectives — what kind of answers one expects — as they will influence the
types of model and analysis that one uses.
(2) Robustness In the physical sciences, models often are based on spe-
cific assumptions about the terms included in the model, and these assump-
tions are based on mechanistic derivations. In the biological sciences, terms
that are included in models do not necessarily have mechanistic derivations,
but are included to give tractable problems. Thus, for example, the logistic
equation often used to model the growth of a single population assumes a
specific form for the growth rate of the population. Many of the results that
can be deduced from a logistic equation model require only that the growth
rate be zero when the population size is zero and when the population size has
a positive value K called the carrying capacity, and that the growth rate be
positive when the population is between zero and the carrying capacity. Ro-
bustness of a model is the property that conclusions drawn from the model are
valid under a less specific set of assumptions. Often one does not know the de-
tails of a biological problem, and one has greater confidence in the conclusions
drawn if changes in detail do not affect the conclusions drawn.
(3) Analysis The mathematical analysis tools and techniques required
vary by model type; for example, stochastic models, which incorporate random
variations, typically require a numerical approach, in which one runs a large
battery of computer simulations where one or more parameters (model inputs)
vary according to a specified probability distribution. For the deterministic
(non-probabilistic) dynamical systems which will form the basis for models
considered in this text, we have three possible approaches: exact, qualitative,
and quantitative. Exact solutions, in which we obtain an explicit formula for a
biological quantity (usually as a function of time), can be obtained only for the
simplest possible types of dynamical systems. We will find them useful mostly
for linearizations, simplifications of complicated models that sketch model be-
havior near special points. Qualitative analysis provides general information
on how a given system behaves, and is typically used to identify what affects
the system in the long term. Quantitative methods use computers to approxi-
3 Castillo-Chávez, as quoted in E. Smith, S. Haarer, J. Confrey, Seeking diversity in math-

ematics education: mathematical modeling in the practice of biologists and mathematicians,

Science and Education, 6: 441–472 (1997).
8 Dynamical Systems for Biological Modeling: An Introduction

mate solutions numerically, and are useful for illustration purposes, or when a
model is too complicated to complete a qualitative analysis, or when we have
good estimates of the parameters involved. In this text we shall show how to
apply all three types of analysis, although qualitative analysis will receive the
most emphasis, as a way to prove how complicated models behave, regard-
less of the exact parameter values. Although the mathematical nature of such
models can generate convincingly real data, it is often important to remember
that, as gross oversimplifications of reality, mathematical models are just as
much rough sketches as the nonmathematical theories with which biologists
are familiar, and qualitative descriptions of behavior befit this notion.
(4) Interpretation Interpreting mathematical results in biological
terms requires a return to the original context of the problem, and is easier
with a well-articulated research question. That is, rather than simply asking,
“What happens if we model the following biological system?” one ought to
ask a specific question that can be addressed with a model — for example,
“What effects do seasonal temperature variations have on the size of this pop-
ulation?” or “How does behavior change based on observed infection levels
affect the eradicability of this disease?” (A general statistical analysis of an
experimental data set, however, may show up trends that shape subsequent
research directions.) Interpretation is also facilitated by a close correspon-
dence between each element of the mathematical model and the elements of
the biological system. The conclusion “this population will go extinct if beta
x is less than 1” is only useful if one can put “beta x” in biological terms —
what do beta and x mean? can we measure them (or at least their product)?
Which might we be able to control, and which are inherent properties of the
biological system?
The basis for an interpretation may be numerical, qualitative, or graphical.
When most parameters (fixed quantities) in a model can be estimated fairly
well, numerical information can provide ranges within which some varying
or controllable quantity causes the biological system to behave in a partic-
ular way, or critical values which cause that behavior to change. Identifying
thresholds between survival and extinction, or between settling down to an
equilibrium level and oscillating forever, or comparing the behavior of two
related but slightly different models, can answer qualitative questions such
as, “Which explanation for this phenomenon better accounts for our obser-
vations?” or “Does the system behave differently if we take into account this
additional factor?” The ready availability of numerous computational tools
also makes it possible to represent results graphically, and a graph can be
an especially compelling, concise statement of a complex result. Graphs can
illustrate both general trends and striking thresholds.
However, just as with algebraic expressions, graphical depictions of math-
ematical information can be so concise that the job of unpacking them into
nonmathematical terms can be significant. The ability of mathematics in all its
forms to represent complicated ideas compactly makes it both powerful and
challenging, and if one considers all the time invested in becoming familiar
 Introduction to Biological Modeling 9

with and manipulating algebraic expressions, it should not be surprising that

unpacking the information contained in graphs requires practice and thought.
For example, Roth4 found that even as apparently simple a graph as that of
population growth rates shown in Figure 1.2 created multiple interpretation
challenges. While a definitive primer in graph interpretation is not the aim of
this text, we shall discuss interpretation of figures we present in later chap-
ters, and also here consider some particular questions regarding the graph in
Figure 1.2 (cf. Figures 2.19 and 2.27 in Section 2.5).

Rate
births deaths

Population size
FIGURE 1.2: A graph comparing birth and death rates vs. population size,
after Roth (2001).

• Note first that the horizontal axis in the graph is not time but population
size. That is, the graph shows how birth and death rates vary depending
on the population, rather than over time. Since many continuous graphs
depict how some quantity changes over time, one must readjust one’s
reading of the graph to reflect what it is that is varying.
• Note second that the vertical axis in the graph gives birth and death rates
(measured in individuals per day, week, or year), and not total births
and deaths. That is, the graph shows how fast births and deaths occur,
as population size grows (varies). The bending downward of the birth
rate curve, for example, as population size increases does not address
what has happened to previous births in the past of any population that
reaches that size. The distinction (and relationship) between population
4 W.-M. Roth (1998), Unspecified things, signs, and “natural objects”: towards a phe-

nomenological hermeneutic of graphing. In S. B. Berenson, K. R. Dawson, M. Blanton,

W. N. Coulombe, J. Kolb, K. Norwood, and L. Stiff (Eds.), Proceedings of the Twentieth
Annual Meeting of the North American Chapter of the International Group for the Psy-
chology of Mathematics Education (Vol. I, pp. 291–297), ERIC Clearinghouse for Science,
Mathematics, and Environmental Education, Columbus, OH, 1998.
10 Dynamical Systems for Biological Modeling: An Introduction

sizes and their rates of change (mathematically, their derivatives with

respect to time) is a fundamental prerequisite of the models we will
develop in the chapters ahead.
• The simpler of the two curves to interpret is the death rate, which ex-
hibits a linear rise, corresponding to a constant per capita death rate (the
total death rate divided by the population size), visible in the graph as
the slope of this line. Since constant per capita death rates due to natural
causes are commonly assumed (see Section 2.1), one can interpret this
line as indicating that there are no density-dependent causes of death in
this system.
• The rise and fall of the birth rate curve, however, suggests that there are
density-dependent forces influencing the birth rate (otherwise this graph
would be a line, too). The decrease in birth rate for larger populations
suggests a factor such as limited resources and/or competition which
reduces reproduction in large populations. (The graph also suggests that
for large enough populations there will be no births at all, which is not
realistic, but see Section 2.4.)
• Ultimately, interpretation of this graph requires the reader to see the
“big picture” depicted here, what one might call the emergent properties
of the graph: what does it communicate? In this case, the superimposi-
tion of the two curves allows us to compare them, to see for what sizes
the population grows and for what sizes it shrinks. Where the birth rate
is greater, the population is growing; where the death rate is greater, it
is shrinking. Our attention should therefore focus on the point(s) where
the two curves intersect, in particular the point where the death rate
catches up with the birth rate. Beyond all the individual details de-
scribed above, the main conclusion this graph was drawn to suggest is
this population level, where (for reasons we shall explore in the chapters
ahead) the population will eventually settle.

Finally, it is also important to acknowledge explicitly the limitations on the

results given: the range within which one’s model is a good enough description
of biological reality, and the extent to which the interpreted results may be
expected to hold true. For example, one of the most frequent simplifications
we shall undertake in our mathematical analyses in this text involves making
nonlinear (complicated) models linear near certain key points. Almost none of
our models will themselves be linear, however. Taylor’s theorem (see Appendix
B) guarantees that linear approximations are reasonable within a very small
operating range, but to replace, say, the graph of the birth rate curve in
Figure 1.2 above with the tangent line at any one of its points would give a
very misleading idea indeed of the relationship between population size and
birth rate. (Recall that the tangent line is the linear approximation to a curve
at that point.)
 Introduction to Biological Modeling 11

(5) Evaluation Evaluating the results given by a particular model re-

quires asking questions such as the following: Are results reasonable? Do they
agree with observations to the extent possible? Do the assumptions made im-
pose too many limitations on the model’s ability to explain or describe the
biological system? The answers to these questions determine whether it is
necessary to change or replace the model. Discussions in the text where evalu-
ation receives special emphasis include those in Sections 2.4, 3.2, 5.1, 5.2, 6.1,
and 7.1.
Note that many of the decisions which need to be made in modeling — like
what features are relevant, what assumptions are reasonable — really become
easier to make only with experience, and the examples in the chapters ahead
provide a beginning. Keep at it!

1.3 Types of mathematical models

Once a research question has been defined and model development begins,
one must decide what type of mathematical model is most appropriate for the
given situation. In this text we shall discuss dynamic models, called in math-
ematics dynamical systems, which track the change of one or more variables
(for us, usually populations) over time (and occasionally over other variables
as well). This discussion therefore leaves out static models, such as statisti-
cal analyses of datasets, which are nevertheless useful for other purposes. We
shall organize our discussion by considering, in turn, the various choices to be
made.5
One recurring issue in modeling is the question, discrete or continuous?
We shall consider this question in three different regards. The terms discrete
dynamical system and continuous dynamical system generally refer to models
that are discrete or continuous in time. Discrete-time models measure the
size of a population at regular, fixed intervals, saying nothing about its size
at moments in between those snapshots. This approach is appropriate for
systems which change or reproduce in distinct generations (e.g., annually, like
many species of fish), or for which data are available at regular intervals,
and results in a model composed of difference equations, as we shall see in
Chapter 2. Continuous-time models depict population growth continuously
over time and are appropriate for systems with ongoing and/or rapid change,
and manifest as differential equations, as we shall see in Chapters 3 and 4.6 For
5 Another accessible and thorough discussion of types and purposes of mathematical

models applied specifically to the epidemiology of sexually transmitted diseases is given

in G.P. Garnett (2002), An introduction to mathematical models in sexually transmitted
disease epidemiology, Sexually Transmitted Infections 78: 7–12.
6 A detailed comparison between a discrete-time model and the continuous-time model

that corresponds to it is outlined in the review exercises at the end of Chapter 3.

12 Dynamical Systems for Biological Modeling: An Introduction

example, suppose we are studying how a certain bird population is affected

by fluctuations in its food source. If the birds are the swallows of Mission San
Juan Capistrano, in California, around which recent development has seriously
depleted nearby insect populations, we might consider a discrete-time model
spanning several years and focus on the number of swallows which return to
the mission on March 19, or the number still at the mission the day before
they migrate south on October 23. However, if we are studying the effects of
seasonal variation in insect availability on a non-migratory bird such as the
Carolina wren, we might instead consider a continuous-time model.
The discrete/continuous question may also arise in two different ways in
defining variables: heterogeneity and population size. If some property, such as
age, spatial location or risk level, varies significantly enough within a popu-
lation to warrant modeling that variation, we must decide whether to incor-
porate the heterogeneity discretely with a stratification or continuously with
a distribution. Stratifications divide a population into groups, each of which
follows similar rules or behaves similarly, and requires defining one state vari-
able for each group. The simplest possible stratification is to define two groups,
such as juveniles and adults, habitat 1 and habitat 2, or high-risk and low-
risk. Continuous distributions introduce an additional independent variable
like age or risk level, and typically lead to partial differential equations, which
are mathematically more complicated. Many human censuses gather data by
age ranges, such as 0–4 years, 5–9 years, 10–15 years of age, etc., and models
built to make use of such data will usually stratify accordingly by age. With
the injunction to simplify in mind, many models stratify into two or a few
groups rather than introduce a second independent variable. For example, a
landmark study on the transmission of gonorrhea in the United States7 fo-
cused on variation in sexual activity levels. Although in practice there is a
wide variation in such levels, the study was able to provide important insights
by stratifying the population into two groups: a low-activity group called the
non-core which includes most of the population, and a small, high-activity
group called the core. However, many applications require detailed model-
ing of spatial or age information, and continuous distributions are important
then. For instance, studies of pattern formation on growing animals (fish,
cats, crocodiles) require keeping careful track of location on the animal in a
continuous way that permits one to see the growth of waves in the patterns.
The state variables which keep track of the sizes of populations and other
dynamic quantities in the model can also be discrete or continuous. At some
level, real populations are always composed of a whole number of individuals
(organisms, molecules, etc.) although of course measures of size such as mass
and volume vary with each individual. Models which keep track of the status
of each individual in a population must by nature have a discrete population
size (even if that size changes with births, deaths, arrivals and departures), but
7 H.W. Hethcote and J.A. Yorke (1984). Gonorrhea transmission dynamics and control.

Lecture Notes in Biomathematics 56. New York: Springer-Verlag.

 Introduction to Biological Modeling 13

models which keep track only of the overall sizes of otherwise homogeneous
groups may allow the variables representing them to vary continuously, espe-
cially if the unit (see Section 1.5) makes this meaningful: biomass measured
in kilograms, or individuals measured in thousands or millions (where each
individual is such a small quantity that the difference between discrete and
continuous is minimal). In general, however, the choice between discrete and
continuous state variables is linked to a more fundamental decision: whether
to make the model deterministic or stochastic.
A deterministic model describes a system in terms of averages: every possi-
ble outcome — dying or not dying, reproducing or not reproducing, recovering
from a disease at any moment from just after infection to years afterward —
occurs, in the proportions corresponding to the relative probabilities of each
event happening. If the average lifetime for a population under study is 30
years, then the corresponding average (per capita) death rate is its reciprocal,
1 1
30 /yr ≈ 0.03/yr, and in a deterministic model precisely 1/30th, or 3 3 %, of the
population will die each year. In some sense, it is more accurate to say that
in a deterministic model, 1/30th of each individual dies per year. This type
of model is appropriate for large populations, where we call on the so-called
Law of Large Numbers, which says that as the number of trials of a random
event (such as whether or not a given individual dies) increases, the actual
experimental probability (here the proportion of the population that under-
goes a certain change) approaches the theoretical probability of the event.
Deterministic models also commonly use continuous state variables.
On the other hand, a stochastic, or probabilistic, model describes all events
in terms of probabilities, and essentially flips a coin or tosses a die for each
individual member of the population, to see whether that member dies or
not, reproduces or not, etc. Consequently, these models use discrete state
variables. A stochastic approach generally requires a little more mathematical
machinery — in particular, familiarity with probability distributions — but is
appropriate in cases where populations are small enough that the Law of Large
Numbers does not apply, or where for some other reason it is important to
couch the model in terms of probabilities: for example, to determine a possible
probability distribution for the outcome of a certain event (like how many
cases an outbreak of disease will cause). Recently, studies have been made of
the structure of social contact networks,8 and in some cases detailed data has
been recorded for contact networks (e.g., SARS outbreaks, and sexual contact
networks in small closed populations); studying how these networks affect
the spread of contact-driven phenomena such as infectious diseases requires
a stochastic approach. Likewise, studies of disease eradication must focus on
the last few infected individuals, where probabilistic effects are important. For
the large-scale study of gonorrhea mentioned earlier, however, a deterministic
8 See, for instance, S.H. Strogatz (2001). Exploring complex networks, Nature 410: 268–

276, and M.E.J. Newman (2003). The structure and function of complex networks, SIAM
Review 45: 167–256.
14 Dynamical Systems for Biological Modeling: An Introduction

approach is not only simpler but has provided significant enough insights to
inform disease control strategies.
In addition, statistical analyses can be especially important for some mod-
els. Two particular calculations of relevance here are sensitivity analysis and
uncertainty analysis. In a sensitivity analysis, a probability distribution of pa-
rameter values is chosen around the parameters’ estimated averages, in order
to determine how strongly the model results depend on each parameter; the
outcome is a ranking of parameters to which some particular model output —
say, the minimum influx of new individuals per year needed in order to sustain
a population which reproduces slowly — is sensitive. For example, one might
find this hypothetical minimum influx to be more sensitive to the existing
population’s death rate than to its reproductive rate. An uncertainty analysis
studies the impact of measurement errors (how far off the estimated param-
eter values are); the outcome is a probability distribution for some model
output, say, the spread of possible values for the required minimum influx
given estimated probability distributions for the birth and death rates.
No single text can hope to make a reader conversant with all of these
types of mathematical models. In this book we shall focus on deterministic
models, with continuous state variables, stratified discretely when necessary
to incorporate heterogeneity, the alternative being continuous variation us-
ing partial differential equations. Chapter 2 will consider models of biological
systems appropriate for discrete dynamical systems (recall that by discrete
we mean discrete in time), while the remaining chapters will consider mod-
els using continuous dynamical systems. In each case, we will begin our study
with simple (linear) single equations, proceed to more complicated (nonlinear)
single equations, and finally consider systems of interacting populations.

1.4 Assumptions, simplifications, and compromises

Modeling, as noted earlier, is an exercise in compromise: we include as
much detail as we can (and must) without losing our ability to analyze the
model. This means that, in developing a model and evaluating the results, we
must identify which simplifying assumptions we make, why we make them, and
how they influence (or limit) the conclusions we draw. It is worth mentioning
some of the trade-offs we make most frequently (in general practice, as well
as in the chapters ahead).
That we simplify at all means that to some extent we must be willing
to give up biological accuracy in exchange for tractability or “analyzability.”
For example, we may assume that the total size of a given population is
constant over time, in order not to have to keep track of it as a separate
variable. Almost always in this text, we shall assume that the amount of
time it takes for any biological process to happen is distributed exponentially,
 Introduction to Biological Modeling 15

the way radioactive decay works (as opposed to, say, a fixed period of time);
although some biological processes do work this way, the real reason we make
this assumption is to make our models be differential equations rather than
something more complicated, like delay or integral equations. Second, at the
same time our willingness to consider reasonably complex models means giving
up exact solutions to the equations in favor of qualitative information about
the solutions: do they eventually taper off toward some equilibrium, or do
they oscillate forever, or do they crash to zero in finite time, or do they do
something stranger? In the case of a quantitative model, we give up exact
solutions and broad, analytical guarantees of possible behaviors in exchange
for quantitative approximations given specific parameter values.9
Third, typically we also make some kind of assumption of homogeneity
of individuals — that, within a given group, they all behave the same way
with regard to any events described in the model. Recall also the discussion of
handling heterogeneity in the previous section; stratification typically includes
this assumption of homogeneity within each class, in exchange for a simpler
type of model. This assumption, as well as the first one above, arises from the
motivation to minimize the number of state variables used (the dimension of
the model).
Fourth, we sometimes give up immediate biological interpretability in re-
turn for tractability — that is, after initially defining a model in biological
terms, we may further simplify it by redefining variables and parameters, in a
way that makes the model mathematically simpler (usually reducing the num-
ber of different variables and/or parameters) but may complicate our ability
to interpret the new quantities in biological terms. For example, one model
for foraging ants10 studied in Section 4.1 begins with five parameters:

N — the total number of ants;

α — the (per capita) rate at which ants find a food site;
β — the (per capita) rate at which ants find a pheromone trail left
by other ants;
σ — the maximum rate at which ants retire from bringing food back
from the site;
K — the number of ants at which the retirement rate reaches 12 σ.

It is determined, however, that the model can be simplified to one involving

only three parameters:

αN K(βN − α) βK 2
a= , b= , c= .
σ σ σ
9 Computational schemes used in numerical analysis, however, are typically shown to

provide approximations that converge to the exact solution as the time-steps involved get
smaller. That is, one can make the approximate solutions as close as one likes to the exact
solution, if one is willing to wait longer for the computer to run its calculations.
10 D.J.T. Sumpter and S.C. Pratt (2003). A modelling framework for understanding social

insect foraging, Behav. Ecol. Sociobiol. 53: 131–144. DOI 10.1007/s00265-002-0549-0.

16 Dynamical Systems for Biological Modeling: An Introduction

The resulting model is simpler to analyze, but the new parameters are more
difficult to interpret in biological terms because the biological information
has been packed mathematically, and must be unpacked following analysis in
order to be interpreted more easily. This process is called rescaling or non-
dimensionalization and is discussed further in Section 1.5 below.
In articulating the simplifying assumptions involved in forming a model,
one should consider the consequences of introducing these “errors.” For ex-
ample, what do we lose from the results of a model if we assume homogeneous
mixing of sediments in a lake, or of a human population? The answers here,
of course, may be very different: in the first case, we lose the ability to dis-
tinguish the effects of sediment build-up as a function of depth, or distance
from the shore, including the resulting distribution of creatures in the lake
that are affected by those sediments (small animals may hide from predators
in the sediment, small animals and plants may even eat it; others, like the
predators, may be hindered by it). In the second case we lose our ability to
determine to what extent some individuals, who come into contact with more
people than most, such as a supermarket checker or a delivery person, may
be instrumental in the transmission of some contact-based process, such as an
infectious disease or the spreading of news and rumors by word of mouth.
There is, finally, another type of simplification which is sometimes made
without any intention of biological justification. In order to make inroads on
a tough problem, we may deliberately oversimplify our model — for instance,
by setting one or more quantities to zero — when the simpler model is easier
to analyze, with the purpose of bringing those results back to the original
problem to inform our intuition and expectations of how the full analysis will
turn out. For example, if there are two processes taking place and one proceeds
much more slowly than the other, we may assume temporarily that the “slow”
variables remain constant. We mention this practice (discussed in Chapter 6)
simply because it can be a useful tool for analyzing models at the edge of
our ability to handle, and because it might not occur to modelers focused
on justifying biologically all mathematical simplifications (the difference here
being that they are only temporary simplifications).
In any case, since models are imperfect descriptions of reality, unless our
main focus really is close fitting of data, our ultimate goal for conclusions
drawn from modeling should be a notion called robustness. This means that the
biological insights obtained from a mathematical model should be independent
of the particular mathematical structure and functions used, to the extent that
we are uncertain what the exact structure or function should be. For example,
as a population grows to fill its habitat, it runs up against the limitations
on the habitat’s resources, which begins to limit the population’s ability to
reproduce. This is true on scales anywhere from a Petri dish to the planet
Earth. This limitation on the birth rate can be described mathematically by
any function which tapers off eventually toward zero after an initial increase
(see, e.g., Figure 1.2). The most common function used for this purpose is
the quadratic that leads to the logistic equation, analysis of which shows that
 Introduction to Biological Modeling 17

the population will eventually level off at a size determined by the resources.
However, other functions of similar shape (that described above) and similar
properties can be used instead, with the same results. Likewise, we might
hope that stratifying a population into three or four age or risk categories will
provide results qualitatively similar to those of a model which stratifies the
population into ten or twenty classes. (We certainly would not like to analyze
the latter model!) Another example is that of threshold quantities, mentioned
above and discussed in Section 4.4. In this way, we justify to a limited extent
the imperfection of our model.
Sometimes models make predictions that are surprising or even alarming.
Such predictions from simple models imply a need for further careful study
of the phenomenon, in order to see if more detailed models make similar
predictions. Predictions which are robust across multiple models are more
likely to be accurate (but no more accurate than the models) and should be
taken seriously.

1.5 Scale, and choosing units

Last in our discussion of what biological modeling is like is a discussion
of scale and units. When one decides to model a certain problem, such as the
variable infectivity observed with HIV, the human immunodeficiency virus, it
is necessary to choose the scale on which to study the problem. Accordingly,
the state variables involved should have appropriate units. At one extreme,
we might choose to focus on the molecular level, in units of nanometers and
seconds — the biochemistry of the immune system’s initial response to the
viral invasion. Moving up a level, we might instead focus on the cellular level,
viewing the interactions of individual virus cells with helper T-cells and in-
fected cells in units of microns, minutes and days. Moving up further, we
might develop a model in the space of a single organ or individual, in units
of centimeters and days to years. Beyond this, we might focus on the inter-
actions of a small collective of individuals, in units of meters and weeks or
months. Finally, at the upper extreme, we might make a model at the level
of an entire population, in units of kilometers and weeks to years. Note that
each jump in scale also effectively uses the lower level as the building block
for the higher level: we model (and count) molecules interacting within a cell,
cells interacting within an organ or individual, individuals interacting within
a collective, and collectives (or individuals) interacting within a population.
This spectrum is illustrated in Figure 1.3.
Arguably the least common, and to some extent newest, of the levels in
this proposed spectrum is that of the collective — a small group within a
population which functions effectively as a unit with regard to the entire pop-
ulation, and within which individuals interact much more with each other
18 Dynamical Systems for Biological Modeling: An Introduction

MOLECULAR CELLULAR ORGAN[ISM] COLLECTIVE POPULATION

FIGURE 1.3: A diagram showing the spectrum of scale levels on which

modeling can take place.

than with others outside the collective. Recent research in mathematical bi-
ology has identified numerous situations in which collectives are important:
for example, individual households in studying infectious diseases. Within a
household, infections tend to spread faster among household members than
between a household member and an individual outside the household. This
is true of human respiratory diseases, as many families know, but also of pests
such as mice, which carry hantaviruses, and insects, which act as vectors for
diseases such as malaria (mosquitoes) and Chagas’ disease (triatomines or
“kissing bugs”). At the population level, it then also becomes reasonable to
make households the unit of measurement, and count infected houses rather
than infected individuals, with recovery coming only when no individuals in
a given household remain infected, or when treatment has eliminated all the
pests in the house.
Another issue related to scale and units is the rescaling procedure men-
tioned in the previous section. Sometimes we find it possible to reduce the
mathematical complexity of a model by reducing the number of variables
and/or parameters. Although the rescaled model can be more difficult to inter-
pret immediately, the resulting simplifications in our calculations are typically
worth it, especially when we can always undo the transformation afterward to
facilitate interpretation. Probably the most common type of rescaling is to re-
define variables relative to benchmark parameters. For example, in the model
of foraging ants mentioned in the previous section (and explored more fully
in Section 4.1), Sumpter and Pratt also rescaled both the state variable E(t),
representing the number of ants exploiting the food site, and the independent
variable t for time, as follows:
E t
x= , τ= ,
K K/σ

so that both x and τ are dimensionless. That is, x gives the number of ex-
ploring ants as a multiple of the benchmark value K, and τ gives the time
as a multiple of the benchmark value K/σ (which is how much time it takes,
on average, for K ants to retire). This is why the process is also called non-
dimensionalization. (The transformation to x in the article also involves elimi-
nation of another state variable for the number of ants not exploiting the food
source by assuming the total ant population constant, another simplifying
technique mentioned in the previous section.)
 Introduction to Biological Modeling 19

Exercises
Consider the following biological systems from a modeling perspective. For
each, determine which type of mathematical model and analysis you believe
would be most appropriate with regard to the criteria developed in this chap-
ter:
• discrete or continuous independent variable(s) (time, age, space),
• discrete or continuous dependent variable (population or quantity),
• type of stratification or continuous distribution of traits, if any,
• deterministic or stochastic,
• quantitative (numerical) or qualitative analysis,
• scale and units.
Defend or justify each of your choices with a sentence. (This exercise is most
useful if discussed in a group setting after each person has written down
his/her choices.)
1. the number of salmon that return to a particular pool to spawn each
year
2. a population of pea aphids whose genetic resistance to infection by fungal
spores varies inversely with their resistance to attack by parasitic wasps
(studied by students in the paper by Smith, Haarer and Confrey)
3. the frequency of a particular allele within each generation of a population
4. the likelihood of extinction of a rare allele (for reasons other than ge-
netic) over many generations of a population
5. the geographic dispersal (movement) of house sparrows within the
United States, starting with the original eight pairs released in the spring
of 1851 in Brooklyn, New York
6. the spread of tall grasses around the shore of a pond from an isolated
cluster
7. seasonal competition for space among different species of trees in a rain-
forest
8. weekly harvesting (removal) of fish in a fish hatchery
9. the growth of a yeast culture in a Petri dish
10. the mixing and growth of bacteria and nutrients in a chemostat
11. the growth of a deer population subject to an annual hunting season
20 Dynamical Systems for Biological Modeling: An Introduction

12. the spread of an epidemic in a large city

13. the role of “super-spreaders” in an epidemic’s growth
14. the growth of a population of blue whales, which become sexually mature
at age 10 years
15. the amount of a given drug remaining in the body over time
16. the kinetics of glucose, insulin and beta cells in the blood of diabetics
17. electrocardiac regulation
18. fluctuations in shark and fish populations in the Mediterranean (the
sharks eat the fish)
19. population density control as a means of eradicating fox rabies
20. comparing two alternative explanations for the mechanisms underlying
cell contraction
21. the concentrations of activator and inhibitor chemicals in pattern for-
mation processes in the hydra, a multicellular water-based creature
22. uptake mechanisms for the diagnostic dye bromosulfophthalein into liver
cells from the blood
23. competition between two species of bird for the same food source and
nesting sites
24. the gradual degradation of the human immune system by HIV
The following three questions provide further practice in graph interpretation.
25. Answer the following questions about Figure 1.2.
(a) What is the significance of the point where the birth rate curve
crosses the x-axis?
(b) What is the significance of the point where the birth rate curve has
a horizontal tangent?
(c) If the population begins at a level greater than that where the two
curves cross, what will happen to its size?
(d) If the population begins at a level less than that where the two
curves cross, what will happen to its size?

26. In what important ways does Figure 1.4 below differ from Figure 1.2?
Consider what will happen to the size of the population if it begins in
each of the three regions marked (a), (b) and (c). (This graph exhibits
something called an Allee effect, which is explored in Section 2.2.)
 Introduction to Biological Modeling 21

Rate
deaths
births

Population size
(a) (b) (c) Graph courtesy M. Tildesley

FIGURE 1.4: Graph for Exer- FIGURE 1.5: Graph for Exer-
cise 26. cise 27.

27. Figure 1.5 shows the distribution of farms in the U.K. during the 2001
outbreak of foot-and-mouth disease (FMD), in terms of the numbers of
cows and sheep on each farm. Color intensity indicates the number of
farms with the given numbers of animals.
(a) Where, on this graph, are the farms with large numbers of cows
but no sheep? Where are the farms with many sheep but no cows?
(b) What does the graph indicate about the distinct types of farms
that exist?
(c) Since there is no treatment for FMD, the epidemic was brought
under control by culling — slaughtering all susceptible animals on
farms within a certain radius of any animal found to be infected.
This wiped out numerous small farms altogether. How might the
small farmers use this graph to argue for a change in control policy?
28. Smith, Haarer and Confrey (1997) studied a group of graduate students
working together in a course on mathematical biology (see footnote 1).
They found that the goals of students from different fields, as well as the
course instructors, who had different backgrounds, differed at times. If
we consider three perspectives from which students using mathematical
models to represent biological systems might come, we might say that:
Those who came with the perspective of an experimental biologist wanted
models to incorporate all their data, and to fit and explain them.
Those who came with the perspective of a theoretical biologist wanted
to use the data obtained beforehand to suggest trends to explore in the
model, to use the model as a means of exploring the effects of particular
phenomena (like genetic variation) across particular systems, i.e., for
many different species and habitats. They saw the model as a virtual
22 Dynamical Systems for Biological Modeling: An Introduction

experiment, whose results become a single datum in the identification

of how general biological systems behave.
Those who came with the perspective of a mathematical biologist wanted
their models to be analyzable, focused only on research-related features
of the system rather than broadly descriptive, and were interested by
mathematical results and their biological significance.
Compare these goals and purposes with each other. To what extent are
they compatible? To what extent would the models produced by each
differ (that is, the models themselves, as opposed to the uses to which
they are put)?
29. Suppose you are investigating the effects of competing predators on the
life cycle of the alewife, a freshwater fish that serves as prey for other
fish such as salmon, trout and bass. Make a list of the five factors you
expect would be most important to include in a mathematical model
for this biological system, and another list of five factors which, while
relevant to the life cycle of the alewife, you would exclude from this
model. Justify both lists biologically.
30. Part of the homogeneity assumption discussed in this chapter is an as-
sumption that when some particular type of contact occurs between
members of two different groups — say, a species of predator and its
prey, or an infectious individual and a susceptible individual — it is
equally likely for the contact to involve any member of each group. In
reality, individuals don’t move about randomly. List five factors which
affect which member of the first group is likely to be the one which in-
teracts with a member of the second group, and indicate some way a
model might take each factor into account.
31. One simplification that most models of biological systems make (and
certainly those in this book) is to assume that events take place in-
stantaneously. That is, infectious contacts between two individuals last
only a moment, and predators and parasites locate and catch their prey
immediately. (The alternative is to have the model remove the two in-
dividuals involved from their respective classes for the time necessary
to make the contact, and after the given delay return the survivor(s)
to those classes, which complicates the model.) What biological justi-
fication can be made for this simplification? In what cases might this
assumption introduce important inaccuracies in the model and its be-
havior?
Chapter 2
Difference Equations
(Discrete Dynamical Systems)

2.1 Introduction to discrete dynamical systems

Many organisms have births, deaths, and other demographic processes that
occur at distinct, regular intervals (see Figure 2.1). Perhaps best-known among
these are semelparous animals that reproduce only once before dying, such as
the Pacific salmon (five species of the genus Oncorhynchus), which battles its
way back upstream from the sea to spawn and die in the river where it was
born. Likewise, the cicada family Cicadidae, which includes some 1500 species
worldwide, includes the genus Magicicada, native to North America, whose
members emerge from their underground hibernation every 13 or 17 years to
mate and die. Analogues also exist in the plant kingdom, such as western North
America’s monument plant (also known as green gentian) Frasera speciosa and
century plant (also known as Parry’s agave) Agavi americana, both of which
live for over twenty years before growing a tall, thin blooming spike which
produces flowers and seeds but saps the plant’s resources to the point that it
dies afterward. Such plants are called monocarpic. Some plants, such as the
carrot Daucus carota, also have fixed life cycles. The carrot and other similar
plants are biennial: at the end of the first year, they store nutrients in the root
that support flowering in the second year, at the end of which they die. On
a smaller scale, there are processes such as ovulation and even neuron firing
which occur discretely at certain set frequencies, rather than continuously.
Finally, human interactions with some populations can also cause discrete cy-
cles, such as harvesting or wildlife population management programs. This
includes situations in which empirical studies necessitate a discrete perspec-
tive on time because the only data available have been gathered at regular
intervals, making everything that happens in between observations equivalent
to one demographic fell swoop.
In cases such as these where it is appropriate to model some biological
quantity or process in terms of the number or amount present at discrete times,
we shall use models which are discrete dynamical systems, that is, systems
which evolve over time through equations (formulas) which tell how to figure
the size of the next generation, based on the size of the present generation.

23
24 Dynamical Systems for Biological Modeling: An Introduction

Photo courtesy U.S. Fish and Wildlife Service, dls.fws.gov Photo courtesy Joel Michaelsen

FIGURE 2.1: The semelparous salmon (left) and the monocarpic century
plant (right) reproduce in distinct generations. Note the different stages of the
blooming spikes on the three century plants.

These equations are called difference equations, since they give the difference in
size between one generation and the next, and first-order difference equations1
have the form yk+1 = f (yk ), which can be read as saying that the size of the
next generation (generation number k + 1 of population y) is a function f of
the present generation (generation number k of population y). In the rest of
this section, we will discuss the simplest type of difference equation and what
its solutions look like. Following that, the next sections in this chapter develop
some tools, both analytical and computational, for studying more complicated
difference equations. Later sections will look at classes of biological systems
that can be studied using discrete dynamical systems, including population
genetics, competition, harvesting, and finally systems involving more than one
quantity or population.

2.1.1 Linear difference equations

The rate of change of some quantity is often proportional to the amount of
the quantity present. This may be true, for example, of the size of a population
with unrestricted growth (say, lab bacteria in a petri dish). Let y(t) represent
the number of members of a population of simple organisms at time t. If we
assume that these organisms reproduce by splitting, and that on average a
fraction a of the members split into two members in unit time, then for a
small period h of time

y(t + h) − y(t) ≈ a h y(t), (2.1)

where the symbol ≈ signifying approximate equality means that the error in
this approximation is small for small h, in the sense that this error divided by
1 First-order difference equations will be our primary focus in this chapter.
 Difference Equations (Discrete Dynamical Systems) 25

h approaches zero as h → 0, i.e., that

y(t + h) − y(t) − a h y(t)
lim = 0.
h→0 h
For now, we neglect the small error and model the population size by the
equation
y(t + h) = y(t) + a h y(t). (2.2)
In this way we can predict y(t + h) given y(t) and a.
If a fraction b of the members reproduce by splitting and a fraction d
of the members die in unit time, then (2.1) would be true with a = b − d.
The constant a may be either positive or negative, depending on whether
b > d (more births than deaths) or b < d (more deaths than births). We will
therefore allow a to designate the constant of proportionality in (2.1) or (2.2),
whether it is positive or negative.
We may rewrite (2.2) in the form

y(t + h) = (1 + ah)y(t), (2.3)

which expresses y(t + h) as a function of y(t). In this chapter we shall think

of h as a fixed time interval, so that if we start at time t = 0 the function
y is defined not for all t but only for t = kh (k = 0, 1, ...). In other words,
we are observing the value of y only at regular intervals, such as every hour,
or every twenty-four hours. If we define tk = kh, the kth observation time,
and let yk = y(tk ), the value observed at that time, then the function y is
described by the sequence of values {yk }. Since tk + h = tk+1 , (2.3) becomes
the difference equation

yk+1 = (1 + ah)yk (k = 0, 1, 2, . . .). (2.4)

In order to simplify the notation, we let r = 1 + ah, so that (2.4) becomes

yk+1 = ryk (k = 0, 1, 2, . . .). (2.5)

By a solution of the difference equation (2.5) we mean an algebraic expres-

sion which gives us values for all the yk (k = 0, 1, . . .). If we know the initial
population size y0 then we can calculate first y1 = ry0 , then y2 = ry1 = r2 y0 ,
etc. Thus we may solve the difference equation (2.5) recursively. The specifi-
cation of the value y0 is called an initial condition. The graph of the solution
is the discrete set of points {(tk , yk ), k = 0, 1, 2, . . .}, but it is customary to
connect these points with line segments to give a continuous graph.
If we return to the bacteria multiplying in the petri dish, and measure time
in multiples of the bacteria’s doubling time, and measure y in multiples of the
original colony size, then Example 1 gives a model for this simple doubling
process. Example 2 gives a variation on this theme, with time steps during
which the colony size increases by 10%, beginning with a colony five times the
size of that in Example 1.
26 Dynamical Systems for Biological Modeling: An Introduction

Example 1.
Solve the difference equation yk+1 = 2yk , with y0 = 1.
Solution: From y1 = 2y0 (the difference equation with k = 0) we see that
y1 = 2y0 = 2. Then y2 = 2y1 = 4, y3 = 2y2 = 8, . . . We may then guess (and
prove by induction) that yk = 2k . This is the solution of the given problem.
Graphing will show the familiar exponential curve. 

Example 2.
Verify that yk = 5(1.1)k is the solution of the difference equation yk+1 =
(1.1)yk , y0 = 5.
Solution: First verify the initial condition: The expression yk = 5(1.1)k with
k = 0 gives y0 = 5(1.1)0 = 5, as desired. Now substitute
 the proposed solution
into the difference equation: yk+1 = (1.1) 5(1.1)k = 5(1.1)k+1 . Thus the
given yk satisfies the given difference equation and initial condition. 

Equation (2.5) fits the general form yk+1 = f (yk ) of a first-order difference
equation; it also belongs to a more specific class of equations of the form

yk+1 = ryk + b, (2.6)

called linear difference equations because the right-hand side f (yk ) is a linear
function of yk . The special case (2.5) where b = 0 is known as the homogeneous
case. These simplest types of difference equation have special names for two
reasons: first, as we shall see below, they are relatively easy to solve outright
(in fact, they are the only kind of difference equation we shall attempt to solve
explicitly); and, second, as we shall see in a later section, they turn out to be
key to understanding the behavior of more complicated difference equations.

2.1.2 Solution of linear difference equations

It is easy to find an explicit formula for the solution of the linear homoge-
neous difference equation with constant coefficients (2.5) for any constant r.
We merely observe that

y1 = r y0 , y2 = r y1 = r 2 y0 , y3 = r y2 = r 3 y0 , . . .

Then it is natural to guess that in general

yk = r k y0 , (2.7)

and it is easy to verify (by induction) that this is correct.

Example 3.
Find the solution of the difference equation yk+1 = −yk , y0 = 1.
Solution: From the formula (2.7) with r = −1 we have yk = (−1)k ; notice
that yk oscillates between positive and negative values. 
 Difference Equations (Discrete Dynamical Systems) 27

Example 4.
Find the solution of the non-homogeneous linear difference equation

yk+1 = −yk + 1, y0 = 1.

Solution: We begin by calculating

y1 = −y0 + 1 = 0, y2 = −y1 + 1 = 1, y3 = −y2 + 1 = 0,

and then conjecture that yk alternates between 0 and 1. To verify the correct-
ness of this conjecture, we need only note that if yk = 1, then yk+1 = 0 and
if yk = 0, then yk+1 = 1. Thus yk does alternate between 0 and 1, which we
may express explicitly as yk = 12 [1 + (−1)k ]. 

It is also not difficult to solve the general linear non-homogeneous difference

equation (2.6) with constant coefficients r and b. We calculate

y1 =r y0 + b
y2 =r y1 + b = r(r y0 + b) + b = r2 y0 + r b + b
y3 =r y2 + b = r(r2 y0 + r b + b) = r3 y0 + r2 b + r b + b, etc.

The formula for the general term is

yk = rk y0 + b 1 + r + r2 + . . . + rk−1 .


(2.8)

By using the formula

1 − rk
1 + r + r2 + . . . + rk−1 =
1−r
for the sum of a geometric series (with r 6= 1), we may write this solution in
the form
1 − rk
 
b b
yk = r k y0 + b = y0 − rk + . (2.9)
1−r 1−r 1−r
We may also consider what happens to solutions of linear difference equa-
tions over long periods of time – in mathematical terms, as k → ∞. Consider-
ing the solution (2.9), we see that if r is large in size, i.e., r > 1 or r < −1, then
rk grows unbounded as k → ∞, and thus yk grows unbounded too. The only
b b
exception occurs when y0 = 1−r , in which case yk is a constant 1−r for all
b
k (cf. (2.9)) and thus approaches 1−r as k → ∞. Small differences, however,
from this equilibrium solution will be magnified for large r (cf. again (2.9)).
If r is instead small in size (between −1 and 1), then rk approaches zero
b
as k → ∞, and in view of (2.9) the solution yk approaches the limit 1−r as
k → ∞ regardless of the initial value y0 .
If r = −1, rk alternates between −1 and 1, and yk does not have a limit
28 Dynamical Systems for Biological Modeling: An Introduction
b
(unless y0 = 1−r ). If r = 1, the formula (2.9) is meaningless, but the formula
(2.8) becomes yk = y0 + rk; thus yk becomes unbounded as k → ∞.
The way the solution of the linear homogeneous difference equation (2.5)
depends on the value of the survival-and-growth term r will be important
when we study qualitative behavior of solutions of difference equations in
Section 2.3. From the formula (2.7) we see that yk → 0 as k → ∞ if |r| < 1
and yk grows unbounded as k → ∞ if |r| > 1. More precisely, if 0 ≤ r < 1,
yk decreases monotonically to zero and if −1 < r < 0, yk oscillates between
positive and negative values in approaching zero. If r > 1, yk increases to +∞
and if r < −1, yk oscillates unboundedly. The “boundary” cases are r = −1,
in which yk oscillates between ±y0 and does not approach a limit, and r = 1,
in which yk is the constant y0 . The essential property we shall need is the
following result.

The solution of the difference equation yk+1 = r yk approaches zero as

k → ∞ if and only if −1 < r < 1.

Example 5.
Find for which values of a every solution of the difference equation yk+1 =
(1 + a)yk approaches zero as k → ∞.
Solution: Every solution approaches zero if and only if |1 + a| < 1, or −1 <
1 + a < 1, or −2 < a < 0. 

2.1.3 Nonlinear difference equations

Note that model (2.5) assumes a constant growth rate independent of pop-
ulation size. This assumption is unlikely to be reasonable for real populations,
except possibly while the population is small enough in size not to be subject
to the effects of overcrowding. Various nonlinear difference equation models
have been proposed as more realistic in the general case, where resource lim-
itations constrain growth. For example, the difference equations
ryk
yk+1 = (2.10)
yk + A

due to Verhulst,2 and the second-order Hill function

ryk2
yk+1 = (2.11)
yk2 + A2

have been suggested as descriptions for populations whose growth rates satu-
rate for large population sizes. Here A is the population size at 50% saturation
2 P. F. Verhulst, Récherches mathématiques sur la loi d’accroissement de la population,

Mem. Acad. Roy. Brussels 18 (1845), 1–38.

 Difference Equations (Discrete Dynamical Systems) 29

(plug in A for yk in (2.10) to see why). For small yk (relative to A), the denom-
inator of the right-hand side of (2.10) and (2.11) is essentially A, so that (2.10)
has yk ≈ Ar yk , similar to the unrestricted growth of the linear equation (2.5).
However, for large yk , A is relatively insignificant, making the right-hand sides
of (2.10) and (2.11) (and thus yk+1 ) approximately equal to r.
Another much-studied example is the logistic difference equation
 yk 
yk+1 = ryk 1 − , (2.12)
K
also introduced by Verhulst, with a growth rate which decreases to zero as yk
approaches the carrying capacity K and which becomes negative for yk > K.
The logistic difference equation should not be taken seriously as a model for
large population sizes as yk+1 is negative if yk > K. Other difference equations
which have been used as models to fit field data are

yk+1 = ryk (1 + αyk )−β (2.13)

and (
ryk1−β , for yk > ǫ,
yk+1 = (2.14)
ryk , for yk < ǫ.
None of the difference equations (2.10), (2.11), (2.12), (2.13), (2.14) are
derived from actual population growth laws. Rather, they are attempts to
give quantitative expression to rough qualitative ideas about the biological
laws governing population growth. For this reason, we should be skeptical
of the biological significance of any deduction from a specific model which
depends on the precise formula for the solution of that model. Our goal should
be to formulate principles which are robust, that is, which are valid for a large
class of models embodying some set of qualitative hypotheses. Therefore, we
shall be more concerned with qualitative properties of solutions of difference
equations than with formulae for solutions. In fact, although we have seen
some examples of linear difference equations for which a solution formula is
available, the solution of nonlinear difference equations like (2.10), (2.11),
(2.12), (2.13), and (2.14) usually cannot be found in general as functions of
the equation parameters. In Section 2.3 we will develop tools for analyzing
the behavior of these more complicated equations.
The most that we can do quantitatively with many nonlinear difference
equations is to calculate solutions numerically by iteration for particular
choices of parameter values, as illustrated by the following two examples.

Example 6
Find the first four terms of the solution of the difference equation yk+1 =
yk (1 − yk ) with y0 = 12 .
Solution: We have y1 = 12 (1 − 12 ) = 14 = 0.25, y2 = 14 (1 − 41 ) = 16
3
= 0.1875,
3 3 39 39 39
y3 = 16 (1 − 16 ) = 256 = 0.152344, y4 = 256 (1 − 256 ) = 0.129135. 
30 Dynamical Systems for Biological Modeling: An Introduction
0.5 1

0.4 0.8

0.3 0.6

0.2 0.4

0.1 0.2

0 1 2 3 4 0 5 10 15 20 25

FIGURE 2.2: Solution to Exam- FIGURE 2.3: A solution to yk+1 =

ple 6. 3.62yk (1 − yk ).

Example 7.
1
Verify that the constant sequence yk = 2 (k = 0, 1, 2, . . .) is a solution of the
difference equation yk+1 = 2yk (1 − yk ).
Solution: If yk = 12 , then 2yk (1 − yk ) = 2( 12 )(1 − 21 ) = 12 . Thus yk = 1
2 satisfies
yk+1 = 2yk (1 − yk ). 

Solutions of difference equations may be calculated and graphed eas-

ily using a computer algebra system such as MapleTM , Mathematica R , or
MATLAB R . Again, although a graph of a solution is technically a set of
discrete points (k, yk ), we shall follow the customary procedure of connecting
the points and presenting the graph as a continuous sequence of line seg-
ments. Figure 2.2, for example, shows the solution of the difference equation
of Example 6 graphed with the aid of Maple; a program for this is given in
Appendix A. Figure 2.3 gives an example which would be more difficult to
calculate by hand, namely 25 terms of the solution of the difference equation
yk+1 = 3.62yk (1 − yk ) with y0 = 0.52. In the next section we shall present a
graphical approach to analyzing difference equations.

Exercises
1. The bacteria E. coli doubles in a little over 20 minutes in good lab
conditions. (For purposes of this exercise, we will assume it takes exactly
20 minutes.)
(a) Write a difference equation that models the growth of an E. coli
 Difference Equations (Discrete Dynamical Systems) 31

colony in a large nutrient dish, with an initial population of y0 =

10000 bacteria, and time measured in increments of 20 minutes.
What is the solution to this equation and initial condition?
(b) How would the equation change if the time between measurements
was one hour instead of 20 minutes? (Consider what happens to
the size of the colony in one hour.)
(c) In light of the discussion in the previous chapter, why would these
models be inappropriate for a colony beginning with a single bac-
terium, or only a few?
(d) Why would these models be inappropriate for tracking the growth
of the colony over a period of months?
(e) Suggest an alternate (nonlinear) model that might be more appro-
priate in tracking the growth of the colony from its inception until
well after it fills the petri dish. Explain in biological terms any
numbers or parameters.
2. Gressel and Segel3 derived the following annual model for the density of
a certain weed species (in weeds per square meter):

wn+1 = F wn + µ,

where wn is the density of the weed in year n, µ accounts for growth

due to mutation from another, closely related species, and F is the ratio
of the fitness of the weed species relative to that of its main competitor
(so, e.g., F > 1 means the weed is more fit than its competitor).
(a) Solve this difference equation to obtain an expression for wn in
terms of µ, F , and the initial condition w0 .
(b) In light of the discussion in this section on difference equations
of this form, what eventually (after many years) happens to the
density of this particular weed if F < 1? Interpret this conclusion
biologically.
(c) If instead F > 1, what happens to the weed density after many
years? Interpret this conclusion biologically.
(d) Which of these conclusions are realistic consequences of the origi-
nal assumptions, and which aren’t? Evaluate this model, and give
a range within which it gives a useful description of the weed’s
growth.
3. Find the first three terms of the solution to yk+1 = yk (1 − yk ), y0 = 34 .
4. Find the first three terms of the solution to yk+1 = yk (1 − yk ), y0 = 41 .
3 J. Gressel and L.A. Segel, The paucity of plants evolving genetic resistance to herbicides:

possible reasons and implications, Journal of Theoretical Biology 75: 349–371, 1978.
32 Dynamical Systems for Biological Modeling: An Introduction
yk
5. Find the first three terms of the solution to yk+1 = yk +1 , y0 = 1.
2
yk
6. Find the first three terms of the solution to yk+1 = 2 +1 ,
yk
y0 = 1.

7. Verify that yk = 1 is a constant solution of the difference equation

yk+1 = yk e1−yk .
8. Verify that yk = 1 is a constant solution of the difference equation
yk+1 = y2y k
k +1
.
∗
9. Verify that if r > 2 the sequence given by
√
(r + 2) + r2 − 4
yk = (k even)
2r
√
(r + 2) − r2 − 4
yk = (k odd)
2r
is a solution of the difference equation yk+1 = yk + ryk (1 − yk ).
∗
10. Show that a constant solution yk = ŷ (k = 0, 1, 2, ...) of a difference
equation yk+1 = g(yk ) must satisfy the relation g(ŷ) = ŷ.
In the following exercises, find the solution of the given difference equa-
tion.

11. yk+1 = (1.1)yk , y0 = 1 15. yk+1 = (1.1)yk − 0.1, y0 = 1

12. yk+1 = −(1.1)yk , y0 = 1 16. yk+1 = −(1.1)yk + 0.1, y0 = 0
13. yk+1 = 12 yk , y0 = 1 17. yk+1 = 12 yk − 12 , y0 = 0
14. yk+1 = 31 yk , y0 = 1 18. yk+1 = 13 yk + 1, y0 = 1

In the following exercises, determine which difference equations have

solutions that approach a limit as k → ∞, and then find the limits in
those cases.

19. yk+1 = −0.2yk 21. yk+1 = 0.2yk + 1

20. yk+1 = − 21 yk 22. yk+1 = − 12 yk − 1
 Difference Equations (Discrete Dynamical Systems) 33

2.2 Graphical analysis

There is a simple graphical method for solving difference equations, called
the cobwebbing method. We shall illustrate the method first by applying it to
the linear homogeneous difference equation

yk+1 = r yk (2.15)

which we have already solved analytically in Section 2.1. The method is also
applicable to difference equations which cannot be solved analytically. It may
be carried out on a computer with the aid of a computer algebra system such
as Maple, Mathematica, or MATLAB. Some of the figures in this section were
produced using Maple, with a program given in Appendix A.
This method begins by drawing in the y − z plane the reproduction curve,
which is the graph of the function on the right-hand side of our difference
equation, and the line z = y, which we shall use for reflection purposes. For
equation (2.15), the reproduction curve is the line z = r y (taken from r yk ).
Next, we mark y0 on the y-axis, and go vertically to the reproduction curve
(meeting it at height ry0 ). The next step is to go horizontally to the line
z = y, meeting it at the point (y1 , y1 ). (Recall y1 = ry0 from (2.15).) Then
we repeat the process, going vertically to the reproduction curve (at height
ry1 ), then horizontally to the line z = y at the point (y2 , y2 ), where y2 = ry1 .
Continuing in the same manner, we reach successively the values y3 , y4 , y5 ,
. . .. The graphic portrayal will have four different cases — r > 1 (Figure 2.4),
0 ≤ r < 1 (Figure 2.5), −1 < r < 0 (Figure 2.6), r < −1 (Figure 2.7) —
corresponding to different relative positions of the reproduction curve and the
line z = y. In each case, the graphical solution illustrates the behavior already

z z=ry z=y z z=y

z=ry

y =r y
1 0
y =r y
2 1
y =r y
2 1

y =r y
1 0

y y
y y y y
0 1 1 0

FIGURE 2.4: Cobwebbing method FIGURE 2.5: Cobwebbing method

for equation (2.15), r > 1. for equation (2.15), 0 < r < 1.
34 Dynamical Systems for Biological Modeling: An Introduction

z z=y z z=y

y0 y0
y y

z=ry
z=ry

FIGURE 2.6: Cobwebbing method FIGURE 2.7: Cobwebbing method

for equation (2.15), −1 < r < 0. for equation (2.15), r < −1.

obtained analytically in Section 2.1. Note that we exclude the case r = 1,

which would make (2.15) yk+1 = yk .
The cobwebbing method may be applied to any difference equation of the
form
yk+1 = g(yk ) (2.16)
using the reproduction curve z = g(y) and the line z = y. It gives informa-
tion about the behavior of solutions and is particularly useful for difference
equations whose analytic solution is complicated.

Example 1.
Apply the cobwebbing method to describe the solutions of the Verhulst equa-
tion
r yk
yk+1 = .
yk + A
ry
Solution: The reproduction curve is z = y+A and its slope is

dz rA
= .
dy (y + A)2

At y = 0 the slope is r/A. If r < A, this slope is less than 1, so the reproduction
curve lies below the line z = y, while if r > A this slope is greater than 1, so the
reproduction curve begins above the line z = y but comes down to intersect
it at y = r − A. If r > A, every solution, regardless of the initial value y0 ,
approaches the limit y∞ = r − A (Figure 2.8), and if r < A, every solution
approaches zero (Figure 2.9). In either case, the limit is an intersection of the
reproduction curve and the line z = y.