MQP Ryl-1213

Project Number: MA-RYL-1213
Math Modeling for Undergraduates
A Major Qualifying Project
submitted to the Faculty
of the
WORCESTER POLYTECHNIC INSTITUTE
in partial fulfillment of the requirements for the
Degree of Bachelor of Science
by
Warren Anderson
Phillip Blake
December 14, 2012
Approved
Professor Roger Y. Lui

Major Advisor
Abstract
This project attempts to write the first draft of a mathematical modeling textbook for un-
dergraduates. We based our information from the course we took under Professor Roger Lui
in B term of 2011. The chapters in this book include: simple pendulum, escape velocity,
Lotka-Volterra equations, extension on Lotka-Volterra equations, populations in competition,
infectious diseases, chemostat, chemical kinetics, enzyme-substrate model, Poincaré-Bendixson
Theorem, and limit cycles. Emphasis of this book is on non-dimensionalization and phase plane
analysis, which includes the study of the stability properties of the steady states. We also use
the MATLAB program pplane8.m to help us plot the phase planes of various models.
1
Acknowledgments
We thank our advisor, Professor Roger Lui, for his advice and help on this project. We also
thank the students in the mathematical modeling class during B term of 2012 for their comments
and feedback on our book. Finally, we thank Professor John Polking of Rice University who
created the pplane8 MATLAB program, which was a tremendous help to us in plotting phase
planes.
2
Contents
1 The Simple Pendulum Model 5

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Phase Plane Analysis with Direction Fields . . . . . . . . . . . . . . . . . . . . . 10
1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Related Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 The Escape Velocity Problem 14

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 The Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Conclusions and Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 The Lotka-Volterra Equations 17

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 The Lotka-Volterra Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Steady States and their Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Phase Plane Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4 Extension on the Lotka-Volterra Equations 27

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 A Logistic Modification of the Lotka-Volterra Equations . . . . . . . . . . . . . 27
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5 Populations in Competition 37
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 The Competition Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.3 Nondimensionalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3
5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6 Models for Infectious Diseases 48

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.2 Models for Infectious Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.3 Steady Sates for the SIRS Model and their Stability . . . . . . . . . . . . . . . . 50
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7 The Chemostat Model 57

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.3 The Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.4 Non-Dimensionalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.5 Steady State Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.6 Phase Plane Analysis and Stability . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.7 Phase Plane Analysis with Direction Fields . . . . . . . . . . . . . . . . . . . . . 66
7.8 Related Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
8 Chemical Kinetics 68
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.2 General Chemical Equations and the Reaction Law . . . . . . . . . . . . . . . . 68
8.3 Rate Law for the Formation of Water . . . . . . . . . . . . . . . . . . . . . . . . 69
8.4 The Graph of d[Hdt2 O] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.4.1 Case 1: [H2 ]o > 2[O2 ]o . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.4.2 Case 2: [H2 ]o < 2[O2 ]o . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
8.5 Autocatalytic Reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
8.5.1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.5.2 Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9 The Enzyme-Substrate Model 83

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
9.2 The Enzyme-Substrate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
9.3 Perturbation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
9.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10 Limit Cycles and the Poincaré-Bendixson Theorem 91

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.2 Poincaré-Bendixson Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.3 Example 1: Scaled Predator-Prey Model . . . . . . . . . . . . . . . . . . . . . . 93
10.4 Example 2: Converting to Polar Coordinates . . . . . . . . . . . . . . . . . . . . 97
10.5 Example 3: The Van der Pol Equation . . . . . . . . . . . . . . . . . . . . . . . 102
4
Chapter 1
The Simple Pendulum Model
1.1 Introduction
As we begin our study of differential equations we will start with a basic model, the simple
pendulum. As with all the models that we will study, we can divide the study of the model
into a few simple sections: Introduction, The Model, The Equations, Phase Plane Analysis with
Direction Fields, and Conclusions. We will delve into what each of these sections entail as we
come to them.
The simple pendulum is defined as a pendulum that moves from a frictionless pivot with a
massless rod connecting the mass to the pivot. If the rod were to have a non-negligible mass
then it would have a different more complicated motion and therefore it would not be a simple
pendulum. The same reasoning applies to the fact that this pendulum has an unbendable rod
instead of some sort of rope or string. The length and shape of the rod are assumed to remain
unchanged in all specific instances of the model. The mass can swing around the pivot point in
a full 360o motion in two dimensional space.
1.2 The Model

In order to derive the equatitons for our model, we will rely on the physical property of balanced
forces. To begin we define a few variables. Let s = arc length which will denote the distance
the pendulum is from its equilibrium. Let θ denote the positive angle in radians between the
rod’s current position and where the rod is positioned when the pendulum is at rest. Let L
denote the length of the rod from the pivot point to the mass.
Using these three properties, we can now develop a useful relationship that will help us later
in developing our model equations.
s θ
= (1.1)
2πL 2π
Or more simply that:
s = θL (1.2)
You can extend this to say that
s̈ = θ̈L (1.3)
5
Figure 1.1: Free Body Diagram for Forces on Simple Pendulum
.
Now to build our equation we must examine the free body diagram of the simple pendulum.
A free body diagram is a drawing which depicts the various forces acting on a system. In
this case we are focusing on the forces which are causing the rod with its mass m to move about
the pivot point.
First and foremost we have that gravity is acting in the downward (negative) direction of
the diagram. This force can be written as −mg where m is the mass of the weight at the end
of the rod and g is the acceleration due to gravity. The tangential force T lays tangent to the
arc length of the path of the pendulum. T can also be referred to as the restoring force which
returns the rod back to its equilibrium position.
We can relate these two forces with a force balance by forming a resultant vector between
the two of them. We can express this relationship in terms of cosines. Remembering that θ is
the angle between the resting position and the current position of the rod:
π T
cos −θ = (1.4)
2 −mg
6
We then rewrite this equation as :
π
T = −mg cos −θ (1.5)
2
Now using the property that:
cos (a − b) = [cos a cos b − sin a sin b]
We are able to rewrite the equation in such a way that

h π π i
T = −mg cos cos(θ) − sin sin(θ) (1.6)
2 2
Now using properties of trigonometric functions it is apparent that cos( π2 ) = 0 and the fact
that sin( π2 ) = 1 we reduce our equation to:
T = mg sin(θ) (1.7)
Now to balance this tangential force as we proposed earlier we will use the properties of
a force balance (more specifically Newton’s 2nd Law F = ma) so find our model’s equations.
As we indicated earlier, in Equation (1.3), we saw that
s̈ = θ̈L
This equation can be rewritten as

ms̈ = mθ̈L (1.8)
Using Newton’s 2nd Law F = ma, we can set Equation (1.8) equal to a force because s̈
represents acceleration. Adding this force to our already derived tangential force we will have a
complete force balance that adds to zero. So our new, almost complete equation can be written
as
mLθ̈ + mg sin(θ) = 0 (1.9)
Finally we can divide by m on both sides, the significance of this operation implies that the
mass of a simple pendulum does not matter when describing it’s motion mathematically. So
our model’s equation can be written as
Lθ̈ + g sin(θ) = 0 (1.10)
More concisely it can be written as

g
θ̈ + sin(θ) = 0 (1.11)
L
1.3 The Equations

To make our model more realistic, we could add air resistance into the equation. In physics
air resistance is assumed to be proportional to velocity. Since we have already determined that
7
θ̈ is acceleration, we can safely say that θ̇ is equivalent to velocity. So our new equation with
resistance can be written as:
g
θ̈ + c θ̇ + sin θ = 0 (1.12)
L
Here, c is the coefficient of friction which accounts for air resistance. From now on we will
refer to Lg as ω02 . The reason this value is squared will prove to be useful shortly.
From here we can attempt to solve our equation for θ(t). To begin we write our system
of given equations which will include initial values θ0 and θ1 for our inital angle and velocity.
We will first solve for the undamped case with no air resistance and we will then solve for the
damped case with air resistance.
Recalling from calculus that sin(x) can be represented as the the following series, we use this
substitution because this is the only way we can solve the model using differential equations,
otherwise the model must be solved numerically.
x3 x5
sin(x) = x − + . . . for small values of x (1.13)
3! 5!
Here this allows us to replace sin(θ) with θ. It is important to note that this is only true for
small values of θ. The reason we do this is now we have a linear equation which can be readily
solved for θ(t) using properties of differential equations.

 θ̈ + ωo2 θ = 0
θ(0) = θ0
θ̇(0) = θ1

We begin by stating the following characteristic equation:
λ2 + ω02 = 0 (1.14)
λ2 = −ω02 (1.15)
q
λ = ± −ω02 = ±ı ω0 (1.16)
This can be written as:

θ(t) = C1 exp(ı ω0 t) + C2 exp(−ı ω0 t) (1.17)
where C1 and C2 are constants.
Using Euler’s formula, we know that:
eıx = cos(x) + ı sin(θ) (1.18)
We can use this to express Equation (1.17) in the following manner:
θ(t) = C1 (cos(ω0 t) + ı sin(−ω0 t)) + C2 (cos(−ω0 t) + ı sin(−ω0 t)) (1.19)

= (C1 + C2 ) cos(ω0 t) + (C1 ı − C2 ı) sin(ω0 t) (1.20)
The arbirtrary constants C1 and C2 can be rewritten as
θ(t) = D1 cos(ω0 t) + D2 sin(ω0 t) (1.21)
8
Now because sin(0) = 0 we use our initial values to show that
θ(0) = D1 ≡ θ0 (1.22)
Taking the derivative of our θ(t) we have the equation that
θ̇ = −D1 ω0 sin(ω0 t) + D2 ω0 cos(ω0 t) (1.23)
Now we can use our other intial value to show that:
θ̇(0) = D2 ω0 ≡ θ1 (1.24)
Finally we can arrive at a precise equation for θ(t) that

θ1
θ(t) = θ0 cos(ω0 t) + sin(ω0 t) (1.25)
ω0
Thus we have a solution for the undamped version.
In the case of damping with the pendulum there are two scenarios to consider. The first
is overdamping where the mass will return to its resting point immediately and not oscillates
back and forth. The other case is where there is underdamping, which is when the mass will
continue to oscillate across the resting point before coming to a stop.
For the damped version we start with the same assumption of θ being small. Thus our three
conditions are the following

 θ̈ + cθ̇ + ωo2 θ = 0
θ(0) = θ0
θ̇(0) = θ1

Recall from MA 2051 Ordinary Differential Equations that this is model can be solved in
the same way that the spring mass system can be solved. The characteristic equation for our
current differential equation is the following
λ2 + 2bλ + ω02 = 0 (1.26)

where 2b = c
Which has characteristic roots:

p p
−2b ± (2b)2 − 4ω02 −2b ± 4b2 − 4ω02
q
λ= = = −b ± b2 − ω02 (1.27)
2 2
We will deal with light or underdamping first. In the case of underdamping the roots are
complex as it is seen that b2 − ω02 < 0
Using the roots of the Equation (1.27), we have the form
λ1,2 = p ± ıq (1.28)
From MA 2051 that with complex roots the solution to the pendulum for theta with respect
of time is of the form:
θ(t) = ept (C1 cos(qt) + C2 sin(qt)) (1.29)
9
This gives us a solution to the underdamped case. Parameters C1 and C2 are dependent on
the specific model and can be derived when given θ0 and θ1
Similarly, we can find the solution for the overdamped case where the roots are not complex
but instead real. Recall that in this scenario, when the pendulum is disturbed from its resting
position it will move back to the rest position in one fluid motion. Recall once again from
MA-2051 the solution form of a 2nd order linear differential equation and we have.
θ(t) = C1 eλ1 t + C2 eλ2 t (1.30)
Here C1 and C2 are constants similar to those in Equation (1.29)
1.4 Phase Plane Analysis with Direction Fields

We will now consider the phase plane analysis for the simple pendulum. A phase plane displays
qualitative data for systems of differential equations. To develop a phase plane we must first
draw what is called a direction field. A direction field gives the general movement of the
particular solutions in the phase plane.
To start let’s take our undamped equation: Equation (1.11)
θ̈ + ω02 sin(θ) = 0
dθ
We then turn this equation into a first order system and let x = θ and y = dt
giving :

d x y
=
dt y −ω02 sin(x)
The steady states for this system are (nπ, 0) , n = 0, ±1, ±2... and so on
The Jacobian for this system evaluated
at the steady states are:
0 1
J1 =
−ω02 cos(nπ) 0
Since cos(nπ) = (−1)n , the characteristic equation of J1 is λ2 + ω02 (−1)n = 0. If n is odd,
then λ = ±ω0 , which means that the steady states are saddles. If n is even, then λ = ±ıω0 ,
which means that the steady states are centers. Below are the direction fields and the phase
plane diagram of the undamped case of the simple pendulum model.
10
Figure 1.2: Direction Field for Undamped Pendulum (Weins)
.
Figure 1.3: Phase Plane for an undamped Simple Pendulum with a rod length of two meters
.
The first order system for the damped case is the following:

d x y
=
dt y −cy − ωo2 sin(x)
The steady states are the same as the undamped case. The Jacobian matrix at the steady
states are:
11

0 1
J2 =
−ωo2 (−1)n −c
The characteristic equation is λ2 + cλ + ω02 (−1)n = 0 and the eigenvalues are:
p
−c ± c2 − 4ω02 (−1)n
λ= .
2
Therefore, if n is odd, then the descriminant is positive and the steady state is a saddle. If n
is even, then the eigenvalues are complex with negative real parts. Therefore, the steady states
are stable spirals. The phase plane for this case is shown below using pplane8.
Figure 1.4: Damped Phase Plane under the same parameters as Figure 1.2
.
1.5 Conclusions
From our study, we have concluded that using slight approximations, the simple pendulum can
be solved using systems of differential equations. There are two general cases of the simple
pendulum, damped and undamped. More specifically there are two different kinds of damping,
overdamping and underdamping. Each of these variations on the model corresponds to different
Jacobian matrices and furthermore they both behave very differently under phase plane analysis.
1.6 Related Questions

1. Give the Jacobian representation for a saddle point in the undamped pendulum model.
Then derive the eigenvalues and eigenvectors from the Jacobian at that point.
12
2. Construct the direction field for the damped simple pendulum. Is there a difference
between the direction fields for the underdamped and overdamped direction fields?
3. Theoretically, is it possible for the pendulum to come to rest at the top of its arc? Propose
conditions that would satisfy this thought experiment and describe the Jacobian at this
point
4. When the roots of a characteristic equation are both equal and real, the system is said to be
critically damped. Describe and plot the critically damped form of the simple pendulum.
13
Chapter 2
The Escape Velocity Problem
2.1 Introduction
Using a few of the methods and tools we have gained throughout the text we are now able to
attack some real world problems. The following problem exemplifies how non-dimensionalization
can reduce the number of variables present and simplify an equation. Even better, the problem
becomes more readily solvable once it is reduced to its basic form.
2.2 The Equations

Consider the equation for the magnitude of attractive force acting between two point masses of
amount M and m separated by a distance r. This equation can be represented as:
GM m
Fr = , (2.1)
r2
where G is the universal gravitational constant.
Now consider a body with mass m being projected from the earth’s surface with initial
velocity V . Letting R be the radius of the earth of mass M and let x(t) be the radial distance
of the earth’s surface at time t. Disregarding air resistance the equation for x is
d2 x GM m
m 2 =− , x(0) = 0, x0 (0) = V (2.2)
dt (x + R)2
Our task now is to find a non-dimensionalized form of this equation and then use that form
to investigate properties of the escape velocity. The escape velocity is the velocity required for
the object with mass m to leave the gravitational field of the earth and be launched into space.
2.3 Solutions
The first part of our solution shall be found from the process of non-dimensionalization. When
we non-dimensionalize we reduce the equation by scaling variables into new parameters. Con-
sidering G , M , R and V we want to chose three new variables , y and τ such that our equation
14
simplifies. Thus we make the following change of variables
x ∗ V V 2R
y= t = τt τ= = (2.3)
R R GM
Note that t∗ is simply a unit carrying term of time and cancels out when undergoing the
process of dimensional analysis. Now, from combining these selections for variables we can
rewrite our equation in the following non-dimensionalized manner.
d2 y 1
2
=− , y(0) = 0, y 0 (0) = 1 = v0 (2.4)
dτ (y + 1)2
The simplifications and reductions have been left as an exercise at the end of the chapter.
Now that we have our non-dimensionalized form we will attempt to look for the value of
the escape velocity. We begin by defining velocity in terms of our new equation. Let v = dy dt
=
velocity, then we have
dv d2 y dv dy dv
= 2 = = v (2.5)
dt dt dy dt dy
Now we rewrite Equation(2.4) as the following:
dv 1
v =− (2.6)
dy (1 + y)2
Now in attempts to isolate v we integrate each side, the left in terms of v and the right in
terms of y. Z v Z y
1
vdv = − 2
dy (2.7)
v0 y(0) (1 + y)
Evaluating the integral we have

v y
v 2 1 1 1
= = − (2.8)
2 v0 1 + y y0
1+y 1+0
v 2 v02

1
− = −1 (2.9)
2 2 1+y
Now we assume that y(t) → ∞ because y(t) is the distance of the object from the earth’s
surface. We assume it is approaching infinity because once the object leaves the earth’s gravi-
tational field it will continuously drift into space.
v2
2
vo 1
= − (2.10)
2 2
Which implies
v02 1
≥ (2.11)
2
However, remember that v0 = 1 from our initial condition. Thus we have
≥2 (2.12)
15
In terms of the original variable you may write as
r
2GM
V ≥ (2.13)
R
Thus the escape velocity can be written as
r
∗ 2GM
V = (2.14)
R
2.4 Conclusions and Questions

We have now shown how to find the escape velocity of an object that is leaving the earth. To
give an idea of how much this quantity actually is, the escape velocity on earth is 11.2 km/s.
It has also been demonstrated that the process of non-dimensionalization can be useful in this
process.
1. Perform the non-dimensionalization discussed in the reading and show how the variables
will cancel after scaling. Afterwards verify the unit of G.
2. At what velocities will the mass create an elliptical or circular orbit around the earth?
16
Chapter 3
The Lotka-Volterra Equations
3.1 Introduction
Differential equations are extremely useful in modeling various biological phenomena. One such
example is the pair of Lotka-Volterra equations, which is a simple representation that models
the population of two species. One species is a predator and the other is its prey.
The Lotka-Volterra equations is one of the oldest predator-prey models. It was proposed
by the biophysicist Alfred Lotka in 1910 to model chemical reactions, and was modified by the
mathematician Vito Volterra in 1925 when he was attempting to explain the oscillating fish
populations he discovered in the Mediterranean. However, the Lotka-Volterra equations can
be applied to many other species as well. The Hudson Bay Company (a famous Canadian fur
trading company) for example observed a similar oscillatory behavior in the populations of the
predatory lynx and its prey, the hare, back in 1840.
3.2 The Lotka-Volterra Equations

The Lotka-Volterra Equations is a system of first-order, nonlinear ODEs. The simplest form of
the system is the following model:
dx
= α1 x − γ1 xy (3.1)
dt
dy
= −α2 y + γ2 xy (3.2)
dt
In the above differential equations, x represents the population of the prey, y represents the
population of the predator, and t represents time. Of course, dxdt
and dy
dt
correspond to the rate
of growth of the two populations with respect to time. The Greek letters α and γ are positive
parameters representing the interaction between the predator and the prey. Table (3.1) shows
what each of the four parameters in the model mean.
The term α1 x means that without the presence of the predator (y = 0), the prey population
will repopulate exponentially without bound. The term −α2 y means that without the presence
of the prey (x = 0), the predator population will decay exponentially. The term −γ1 xy means
17
Table 3.1: Description of Parameters in the Lotka-Volterra Model
Parameter Description Units
α1 Represents the reproduction rate of the prey. The 1/time
greater α1 is, the more rapidly the prey reproduces.
γ1 Represents the death rate of the prey due to the pres- 1/time
ence of the predator. The greater γ1 is, the greater
the death rate of the prey is due to predation and the
more effective the predator is killing the prey.
α2 Represents the death rate of the predator. The 1/time
greater α2 is, the greater the death rate of the preda-
tor is.
γ2 Represents the reproduction rate of the predator. 1/time
The greater γ2 is, the more rapidly the predator re-
produces and the more effectively the prey is able to
nourish the predator.
that the prey population will decline due to the presence of both species. This is because the
predator kills the prey and because a greater prey population is easier to hunt than a smaller
prey population. Finally, the term γ2 xy means that the predator population will increase due
to the presence of both species.
Unfortunately, as with all mathematical models, the Lotka-Volterra equations cannot be used
for every situation. This is because there exist numerous assumptions that are made about the
two species that restrict the applications of the model. These assumptions are in place so that
the model does not become too complex. If the Lotka-Volterra model could be applied for every
predator-prey relationship, it would be so complex that an analysis of it would be exceedingly
(if not impossibly) difficult! In fact, such a model would be so hard to understand and analyze
that it would be for all practical purposes, useless. Indeed, a mathematical models exist so that
we can understand them and apply them to real-world phenomena.
One assumption is that the survival of the predator entirely depends on the presence of
one species, the prey. Another assumption is that without the presence of the predator, they
prey’s population will grow exponentially without bound. A third assumption is that the prey
is always able to find ample food and shelter. A fourth assumption is that both species roam
about randomly in their habitat and are uniformly distributed. There are other assumptions,
but these ones are the most important.
As you can see, these assumptions restrict the predator-prey pairs that may be appropriately
modeled using the Lotka-Volterra equations. For example, foxes prey on rabbits, but their
survival does not strictly depend on rabbits. Foxes are omnivores, and diet on mammals,
amphibians, reptiles, birds, insects, and vegetation. Therefore, even though the fox population
may suffer without the existence of rabbits, the species would not become extinct without them,
as the Lotka-Volterra equations would suggest. (To see this, substitute x = 0 into the model.
Notice that dy
dt
would become negative.) On the other hand, the Canadian lynx that the Hudson
Bay Company observed almost exclusively feeds on the horseshoe hare, and so the Lotka-Volterra
18
equations would be much more appropriate to model the lynx and hare population.
3.3 Steady States and their Stability

We are now ready to preform a phase plane analysis on the Lotka-Volterra equations! We can
begin by rearranging Equations (3.1) and (3.2) as such:
dx
= (α1 − γ1 y)x (3.3)
dt
dy
= (−α2 + γ2 x)y (3.4)
dt
It is very easy now to identify the nullclines for the system by equating each differential
equation to zero and solving for x and y:
α1
f-nullclines: x = 0 and y = γ1
α2
g-nullclines: y = 0 and x = γ2
Then, we can find the steady states by finding the points at which the f-nullclines intersect
with the g-nullclines. They are:
(x1 , y1 ) = (0, 0) and (x2 , y2 ) = ( αγ22 , αγ11 )
Now we can calculate the Jacobian of the system by taking the partial derivatives with respect
to x in the first column, and respect to y in the second column. We can also evaluate the
Jacobian at the two steady states:

α1 − γ1 y −γ1 x
J(x, y) =
γ2 y −α2 + γ2 x

α1 0
J(0, 0) =
0 −α1
0 − γ1γα2 2

α2 α1
J , = γ2 α1
γ2 γ1 γ1
0
Using the Jacobians, we can calculate the eigenvalues and eigenvectors for each steady state.
For the steady state (0,0):
det(J(0, 0) − λI) = (α1 − λ)(−α2 − λ) = 0

∴ λ1 = α1 > 0, λ2 = −α2 < 0
19
Solving for eigenvectors v1 and v2 :
(J(0, 0) − λ1 I)v1 = 0 (J(0, 0) − λ2 I)v2 = 0

0 0 v11 0 α1 + α2 0 v21 0
= =
0 −α2 − α1 v12 0 0 0 v22 0

0 0 v21 (α1 + α2 ) 0
= =
−v12 (α1 + α2 ) 0 0 0

1 0
∴ v1 = ∴ v2 =
0 1
Note that any value (except 0) for v11 and v22 could have satisfied the equations, but 1 is chosen
for simplicity. Since λ2 < 0 < λ1 , (0,0) is a saddle point. However, how do we determine if
the solutions are stable (approach the steady state) or unstable (avoid the steady state)? The
answer is quite simple: if the eigenvalue is positive, the solution is unstable along its corre-
sponding eigenvector; if the eigenvalue is negative, the solution is stable along its corresponding
eigenvector. Therefore, solutions are unstable for λ1 since it is a positive eigenvalue, and those
solutions lie on the vector v1 = (1, 0). In other words, the solutions avoid the origin along
the positive x-axis. In addition, the solutions are stable for λ2 since it is a negative eigen-
value, and those solutions lie on the vector v2 = (0, 1). Thus, the solutions approach the
origin along the positive y-axis. Using those two facts, it is easy to draw solutions near
(0,0). However, we must consider the other steady state before drawing the whole phase plane
diagram. Figure (3.1) shows the first quadrant of the phase plane near the origin. One of the
g-nullclines is the x-axis in orange and one of the f-nullclines is the y-axis in pink. Notice that
solution curves avoid the origin along the positive x-axis, and approach the origin along the
positive y-axis as discussed above.
For the steady state ( αγ22 , αγ11 ):

α2 α1
det J , − λI = λ2 + α1 α2 = 0
γ2 γ1
√ √
∴ λ1 = α1 α2 i, λ2 = − α1 α2 i
We could find the eigenvectors corresponding to the eigenvalues, but this would not be beneficial
in this case because the eigenvectors would be on imaginary axes while our phase plane is on
real axes. However, there is an alternative method that we can use to determine the direction
of the solution curves. First note that the solutions near ( αγ22 , αγ11 ) are centers because
λ1 and λ2 are pure imaginary numbers. To determine if the solutions are going clockwise or
counterclockwise around the steady state, we may choose any point near it and determine the
direction of the solution using the the system of differential equations. For example, the point
( α1 γγ21+α
γ2
2 γ1 α1 γ2 +α2 γ1
, γ1 γ2 ) is located northeast of ( αγ22 , αγ11 ). Note that αγ11 + αγ22 = α1 γγ21+α
γ2
2 γ1
, so the point
α2 α1 α1 α2
chosen is the sum of ( γ2 , γ1 ) and ( γ1 , γ2 ). Substituting this point into Equation (3.3) yields:

dx α1 γ2 + α2 γ1 α1 γ2 + α2 γ1 α2 γ1 α1 γ2 + α2 γ1
= α1 − γ1 = −
dt γ1 γ2 γ1 γ2 γ2 γ1 γ2
20
Figure 3.1: This is what the phase plane of the Lotka-Volterra equations looks like close the
the origin, a saddle point. Notice that v1 lies on the positive x-axis and v2 lies on the positive
y-axis.
21
dx
Figure 3.2: Since dt
< 0 in the NE quadrant, the solution curves must go anticlockwise around
the steady state.
Recall that all of the parameters are positive. Therefore, α1 γγ21+α

γ2
2 γ1
is positive. Furthermore, the
quantity inside the parenthesis is always negative. We can conclude then, that dx dt
< 0 northeast
α2 α1 α2 α1
of ( γ2 , γ1 ) and so solutions go counterclockwise around ( γ2 , γ1 ). Note that if dx dt
> 0, then
α1 γ2 +α2 γ1 α1 γ2 +α2 γ1
the solutions would be clockwise. We could have alternatively substituted ( γ1 γ2 , γ1 γ2 )
into the dy dt
equation to yield the same result. If dy
dt
> 0 (which it is), then the solutions go
dy
counterclockwise. If dt < 0, then the solutions would be clockwise. Take a look at Figure (3.2)
for clarification.
3.4 Phase Plane Analysis

If you have a computer algebraic system such as Maple, Mathematica, or a TI-89, you can draw
implicit plots of the solution curves. In order to do this, you need to derive the equation for
dy
dx
, separate the variables, and solve implicitly for x and y by integrating. The steps are shown
below:
22
dy (−α2 + γ2 x)y
= (3.5)
dx (α1 − γ1 y)x
(α1 − γ1 y) (−α2 + γ2 x)
dy = dx (3.6)
y x
Z Z Z Z
α1 1
dy − γ1 dy = −α2 dx + γ2 dx (3.7)
y x
α1 ln(y) − γ1 y = −α2 ln(x) + γ2 x + C (3.8)
ln(y α1 ) + ln(xα2 ) = γ2 x + γ1 y + C (3.9)
ln(xα2 y α1 ) = γ2 x + γ1 y + C (3.10)
It is left as an exercise to show that the domain of C is C ≤ α1 ln( γα11e )+α2 ln( γα22e ). Note that
when all the parameters are equal, the domain of C simplifies to C ≤ − 2α1 . A limitation of
many computer algebraic systems is that they have difficulty plotting graphs with steep vertical
slopes and slopes that change dramatically, so you may have problems plotting solution curves
when C is much smaller than its maximum value.
In order to plot the graphs into Maple, use the commands:
> with(plots);
> implicitplot(ln(x0.25 · y 0.25 ) = 0.25 · x + 0.25 · y − 0.54, x = 0..2, y = 0..2, gridrefine = 3);
In the example above, γ1 = γ2 = α1 = α2 = 0.25 and C = −0.54. (Since all the parameters
are equal the domain of C is C ≤ − 2α1 , so C ≤ − 0.5.) The first parameter for implicitplot
is the equation to be plotted. The second and third parameters are the x and y values that
Maple will evaluate the plot. After all calculations Maple will automatically zoom for best fit,
so the domain and range may be different than the values you put here. The fourth parameter
is optional. Gridrefine tells Maple to compute more points on the plot for a graph with a higher
resolution. Although the plot will usually look better with higher gridrefine values, computation
time will increase. Do not use a value higher than 3 unless you know your computer can handle
the workload. Figure (3.3) shows the plot Maple produces after the two commands above are
executed. Note that the origin is not displayed on the graph!
Figure (3.4) a phase plane diagram for the Lotka-Volterra equations. The x-axis is the
prey population and the y-axis is the predator population. We are only interested in the first
quadrant because the population of the species cannot be negative! The f-nullclines are in pink,
the g-nullclines are in orange, and the solution curves are in blue. Values of 41 have been used
for all the parameters α1 , α2 , γ1 ,and γ2 . Therefore the steady states are (0,0), which is a saddle,
and (1,1), which is a center. There are several arrows on the diagram as well, which represent
the direction the solution curves go in. Notice that on the f-nullclines, the solution curves are
always vertical because dx dt
= 0 and that on the g-nullclines, the solution curves are always
dy
horizontal because dt = 0.
If we analyze the phase plane diagram, we will encounter a few surprising results. One
interesting phenomena is that the steady state levels for both species are independent of their
growth and mortality rates. Rather, the steady state level of the prey depends on the ratio αγ22
and the steady state level of the predator depends on the ratio αγ11 . What this means is that in
order for a steady predator population to exist, the prey population must just suffice so that
23
Figure 3.3: Maple produces this plot for the equation ln(x0.25 y 0.25 ) = 0.25x + 0.25y − 0.54
Figure 3.4: This is the phase plane diagram for the Lotka-Volterra equations, where α1 = α2 =
γ1 = γ2 = 0.25. Steady states are (0,0) and (1,1).
24
the growth rate of the predator, γ2 x, is equal to the predator mortality rate α2 . Similarly, in
order for a steady prey population to exist, the predator population must just suffice so that
the growth rate of the prey, γ1 x, is equal to the prey mortality rate α1 .
Another surprising result is that the steady state ( αγ22 , αγ11 ) is a center, and not a spiral. Thus,
the predator and prey populations will never reach an equilibrium point in which the population
of both species remains a constant (unless the initial population of the species happen to be
exactly on that steady state, which is extremely unlikely). Rather, the population of both
species will oscillate indefinitely.
A third peculiar result is that the Lotka-Volterra model rarely predicts that a species will
go extinct. Even when one population is very close to either the x axis or the y axis, the model
predicts that the population of the species will rebound to safe levels. To see this, just look at
the phase plane diagram, and trace the solution curves near the axes. Of course, due to the
oscillatory nature of the model, the species will continue to phase in and out of endangerment.
Lastly, notice that the model ignores the fact that population must be in whole numbers.
When a solution is exceptionally close to either the x or y axis so that only a fraction of a
species exist, the model always predict that the species will be able to repopulate. This is of
course unrealistic. In reality, if a population of a species is less than one, then it is extinct.
It is very important when working with mathematical models to understand the strengths and
limitations of them. One weakness of the Lotka-Volterra model is that it does not realistically
predict the population of predator-prey ecosystems when one or both species are very close to
extinction.
3.5 Conclusion
In this chapter, the Lotka-Volterra model was introduced. the Lotka-Volterra model is a system
of differential equations that models the oscillatory behavior of a predator-prey ecosystem. The
system has two equations. One equation models the rate of change of the prey population, and
the other equation models the rate of change of the predator population. Although the system
is nonlinear, a phase plane diagram can be produced so that we can analyze the behavior of
the system. Doing so produces a few surprises. One result of the phase plane analysis is that
there are only two steady states, one in which the population of both species becomes extinct
and another in which the populations of both species circle around. However, because the
origin is a saddle the model rarely predicts that both species will actually become extinct, and
so no matter what the populations of both species are, they will oscillate around the other
steady state. Another result of the phase plane analysis is that the neutral stability of both
populations do not depend on the populations of either species. Rather, the stability depends
on the parameters α1 , α2 , γ1 , and γ2 .
3.6 Problems
1. Consider the following symbiotic model in which two species help each other grow. Both
species grow logistically without the presence of the other species. When the other species
is present, however, the population growth increases faster than logistically. The symbiotic
25
model is:
dx
= α 1 x − β 1 x2 + γ1 xy
dt | {z } |{z}
logistic growth of x without y symbiotic growth
dy
= α2 y − β2 y 2 + γ2 xy
dt | {z } |{z}
logistic growth of y without x symbiotic growth
(a) Find all nullclines of the system.

(b) Identify all steady states.
(c) Find the Jacobian of the system at each steady state.
(d) Classify each steady state as a spiral, center, saddle, etc. and determine the stability
at each steady state, assuming β1 γ2 > β2 γ1 .
2. In Section 3, we discovered that the steady state ( αγ22 , αγ11 ) is associated with pure imaginary
eigenvalues. If the system were liner we could conclude that the steady state is a center,
but since the Lotka-Volterra equations are nonlinear the process during linearization could
have perturbed the eigenvalues so that they are not purely imaginary (i.e. λ = a + bi).
Therefore, we must consider the possibility that the steady state is a spiral.
(a) Show that Equation (3.10) is equivalent to:
y α2 e−γ1 y = Kx−α2 eγ2 x
Where K is an arbitrary constant.
(b) On the nullcline x = αγ22 observe that y α2 e−γ1 y = constant. Graph the function
f (y) = y α2 e−γ1 y and use your graph to demonstrate that f (y) = constant can have
at most two solutions for any given constant K.
(c) Conclude that the trajectory cannot be a spiral. (Hint: Consider how many times
f (y) intersects the line x = αγ22 .
3. The domain for C in Equation (3.10) is C ≤ α1 ln( γα11e ) + α2 ln( γα22e ). In this exercise, you
will verify this result.
(a) Note that Equation (3.10) is equivalent to:
C(x, y) = ln(xα2 y α1 ) − γ2 x − γ1 y (3.11)
Find the critical point, (a, b), of this equation by finding the solutions of ∂C
∂x
= 0
∂C
and ∂y = 0. Notice that the critical point is a steady state for the Lotka-Volterra
equations!
(b) Let Q = Cxx (a, b), R = Cxy (a, b), and S = Cyy (a, b). The critical point (a, b) is a
maximum of C(x, y) if Q < 0 and QS − R2 > 0. Use this fact to show that the
critical point is a maximum.
(c) Evaluate C at (a,b) to find the maximum value of C. Verify that:
α1 α2
C(a, b) = α1 ln( ) + α2 ln( )
γ1 e γ2 e
26
Chapter 4
Extension on the Lotka-Volterra

Equations
4.1 Introduction
In the previous chapter, a system of differential equations was introduced to model the popula-
tions of a predator-prey ecosystem. One disadvantage of this model, however, is the assumption
that they prey population increases exponentially without the presence of the predator. In
many cases, this assumption is unrealistic. Usually the prey population increases logistically
rather than exponentially in the absence of the predator. This is because natural resources
such as a limited food supply restrict population growth. In this chapter, we will modify the
Lotka-Volterra equations so that the prey population experiences logistic population growth.
4.2 A Logistic Modification of the Lotka-Volterra Equa-

tions
In 1838, Pierre-François Verhulst proposed the following logistic equation to model the popula-
tion growth of a species:

dP P
= rP 1 − (4.1)
dt K
The differential equation above has the solution:
P0 K
P (t) = (4.2)
P0 + (K − P0 )e−rt
In Equation (4.2), P0 is the initial population of the species and P (t) is the population at time
t. Of course, P0 , P (t), and t must be positive. To understand what the constant K is, we may
take the limit of Equation (4.2) as t approaches infinity, as such:
P0 K P0 K
lim P (t) = lim −rt
= =K (4.3)
t→∞ t→∞ P0 + (K − P0 )e P0
27
1000
Figure 4.1: Three logistic equations, P (t) = 10+90e−rt , with r values 0.05, 0.1, and 0.2. Here, we
can see the effect r has on P (t). The greater r is, the faster P (t) approaches K.
It is clear now that the value of K is the population the species will tend to as time becomes
large. K is called the carrying capacity of the species. Finally, to determine what the value of
r does, we can plot several graphs of P (t) with different r values. In Figure (4.1), P0 = 10 and
K = 100 for all three graphs, and r varies. Notice that for all the graphs, the initial population
is 10 and the carrying capacity is 100. The difference among the graphs is how quickly the
population approaches the carrying capacity K. The greater r is, the faster the population
approaches the carrying capacity.
We now have all the tools that we need to understand the modified Lotka-Volterra equations.
The system of differential equations is:
dx x
= α1 x 1 − − γ1 xy (4.4)
dt K
dy
= −α2 y + γ2 xy (4.5)
dt
Comparing Equation (4.4) with Equation (3.1) and Equation (4.5) with Equation (3.2), you
may notice that Equations (4.5) and (3.2) are equivalent. It would be incorrect to apply the
logistic equation for the predator because the predator’s population growth already depends
on x, which grows logistically already. Table (3.1) in the previous chapter describes all of the
parameters in the logistic Lotka-Volterra model except for K, which is the carrying capacity of
the prey population. Since K measures a positive quantity, it is unitless.
28
Our end goal is to preform a phase plane analysis on the logistic Lotka-Volterra equations. The
first step should always be finding the nullclines of the system by equating dx
dt
= 0 and dy
dt
= 0.
The f-nullclines are:
dx x
= 0 = α1 x 1 − − γ1 xy (4.6)
dt K
α1 x
0 = x α1 − − γ1 y (4.7)
K
α1 (K − x)
0=x − γ1 y (4.8)
K
α1 (K − x)
x=0 or 0= − γ1 y (4.9)
K
α1 (K − x)
y= (4.10)
γ1 K
The g-nullclines are:

dy
= 0 = −α2 y + γ2 xy (4.11)
dt
0 = y(−α2 + γ2 x) (4.12)
y=0 or 0 = −α2 + γ2 x (4.13)
α2
x= (4.14)
γ2
The next step is to find the steady states, which are easily obtainable by finding the intersection
of the f-nullclines with the g-nullclines. (There are no steady states when two f or two g-nullclines
intersect each other!) We can find these either graphically or algebraically. The steady states
are:
(x1 , y1 ) = (0, 0) (4.15)

 
α α1 K − αγ22
α2 α1 α1 α2

2
(x2 , y2 ) =  ,  = , − (4.16)
γ2 γ1 K γ2 γ1 γ1 γ2 K
α1 (K − x)
(x3 , y3 ) = (K, 0) since y= =0 implies that x=K (4.17)
γ1 K
To determine the stability of the steady states, finding the Jacobian of the system will be helpful.
The Jacobian can be found by taking the derivative with respect to x in the first column and
taking the derivative with respect to y in the second column. The Jacobian is:
α1 − 2αK1 x − γ1 y

−γ1 x
J(x, y) = (4.18)
γ2 y −α2 + γ2 x
29
For the steady state (x1 , y1 ), the Jacobian simplifies to:

α1 0
J(0, 0) = (4.19)
0 −α2
The eigenvalues at (x1 , y1 ) are:
det(J(0, 0) − λI) = (α1 − λ1 )(−α2 − λ2 ) = 0 (4.20)

∴ λ1 = α 1 > 0 and λ2 = −α2 < 0 (4.21)
Notice that J(0, 0) is a diagonal matrix, so the calculations above actually are not necessary
since the eigenvalues are the elements on the diagonal. Since λ2 < 0 < λ1 , (0,0) is a saddle
point. To determine the direction of the solution curves, we can calculate the eigenvectors for
each eigenvalue:
(J(0, 0) − λ1 I)v1 = 0 (J(0, 0) − λ2 I)v2 = 0 (4.22)

0 0 v11 0 α1 + α2 0 v21 0
= = (4.23)
0 −α2 − α1 v22 0 0 0 v22 0

0 0 v21 (α1 + α2 ) 0
= = (4.24)
−v12 (α2 + α1 ) 0 0 0

1 0
∴ v1 = ∴ v2 = (4.25)
0 1
Since λ1 > 0, solutions avoid the steady state (0,0) along its corresponding eigenvector. Its
eigenvector is (1,0), which goes along the positive x-axis. Therefore, solutions avoid (0,0)
along the positive x-axis. Also, since λ2 < 0, solutions approach (0,0) along its corresponding
eigenvector. Its eigenvector is (0,1), which goes along the positive y-axis. Thus, solutions
approach (0,0) along the positive y-axis. Notice that the stability of the origin is the same
for this model as it was for the Lotka-Volterra equations in the previous chapter.
Now we can find the stability of the next steady state, (x2 , y2 ). We begin as usual by finding
its Jacobian:
−γ α
!
−α1 α2 1 2
γ1 K γ2
J(x2 , y2 ) = α1 (γ2 K−α2 ) (4.26)
γ1 K
0
We could continue to find the eigenvectors for (x2 , y2 ), but the computations would be
tedious. There is, however, a workaround. We can calculate the trace and the determinant
of the Jacobian instead. Doing this, we obtain:
−α1 α2
tr(J(x2 , y2 )) = <0 (4.27)
γ1 K
α1 α2 (γ2 K − α2 )
det(J(x2 , y2 )) = >0 (4.28)
γ2 K
Recall that all of the parameters and variables must be positive. Therefore, tr(J(x2 , y2 )) < 0.
In order for y2 > 0, K > α2 /γ2 . This implies that det(J(x2 , y2 )) > 0. Since tr(J(x2 , y2 )) < 0
30
and det(J(x2 , y2 )) > 0, (x2 , y2 ) must be a stable. Further analysis shows that tr(J(x2 , y2 ))2 <
4 det (J(x2 , y2 )), and so the point (x2 , y2 ) must be a stable spiral. We now calculate the
Jacobian of the last steady state, (x3 , y3 ):

−α1 −γ1 K
J(x3 , y3 ) = (4.29)
0 γ2 K − α2
This time, it is not very difficult to find the eigenvalues for (x3 , y3 ):
det(J(x3 , y3 ) − λI) = (−α1 λ1 )(γ2 K − α2 − λ2 ) = 0 (4.30)

∴ λ1 = −α1 < 0 and λ2 = γ2 K − α2 > 0 (4.31)
Observe that the eigenvalues are the elements on the diagonal since J(x3 , y3 ) is an upper trian-
gular matrix. Since λ1 < 0 < λ2 , (x3 , y3 ) is a saddle. To determine the direction of the solution
curves, we calculate the eigenvectors for each eigenvalues, as shown below:
det(J(x3 , y3 ) − λ1 I)v1 = 0 det(J(x3 , y3 ) − λ2 I)v2 = 0 (4.32)

0 −γ1 K v11 0 −α1 − γ2 K + α2 −γ1 K v21 0
= = (4.33)
0 γ2 K − α2 + α1 v12 0 0 0 v22 0

−v12 γ1 K 0 v21 (−α1 − γ2 K + α2 ) − v22 γ1 K 0
= = (4.34)
v12 (γ2 K − α2 + α1 ) 0 0 0
−γ1 K
1
∴ v1 = ∴ v2 = α1 +γ2 K−α2 (4.35)
0 1
Since λ1 < 0, solutions approach (x3 , y3 ) on the ray containing v1 . Also, since λ2 > 0, solutions
avoid (x3 , y3 ) on the ray containing v2 .

In the previous chapter, we were able to find an implicit equation for the solution curves.
For the logistic model, however, this is not possible because solving the differential equation
algebraically:
dy −α2 y − γ2 xy
=
dx α1 x(1 − x/K) − γ1 xy
is extremely difficult. In fact, not even Maple can solve the differential equation explicitly! It
is, however, quite possible to draw a phase plane without solving the differential equation. To
do this, first we need to plot all of the nullclines and the steady states. Then, it is easy to
draw direction fields on the nullclines. All direction fields on the f-nullclines are vertical since
dx
dt
= 0, and all direction fields on the g-nullclines are horizontal since dy dt
= 0. To determine
whether the direction fields are going up, down, left, or right, simply determine the sign of dy dt
for the f-nullclines or dx
dt
for the g-nullclines. Only when a nullcline crosses a steady state may
the sign of a direction field change. In order for a direction field to change sign, the derivatives
31
must change sign. The only possible points where the derivatives can change sign is when the
derivatives are zero, which occurs at the steady states.
For example, consider the g-nullcline x = α2 /γ2 . The direction fields along this nullcline
must be horizontal since dy dt
= 0. The steady state (x2 , y2 ) is on the nullcline. The direction
fields below (x2 , y2 ) go rightward. This is because dx dt
= α1 x(1 − x/K) − γ1 xy > 0 for all points
on the nullcline below (x2 , y2 ), assuming K > α2 /γ2 . In particular, (α2 /γ2 , 0) is one point on
the nullcline that is below (x2 , y2 ). At (α2 /γ2 , 0):

dx α1 α2 α2 /γ2
= 1− >0 assuming K > α2 /γ2
dt γ2 K
Recall that we may make this assumption on K because K must be greater then α2 /γ2 in order
for y2 > 0. Furthermore, the direction fields above (x2 , y2 ) go leftward. This is because dx dt
<0
for all points on the nullcline above (x2 , y2 ).
Figure (4.2) shows the nullclines, steady states, and direction fields on the nullclines. Note
that we are only concerned with the first quadrant since x and y are always nonzero. Using the
fact that (0, 0) and (K, 0) are saddle points and (x2 , y2 ) is a stable spiral, it becomes very easy
to complete the phase plane diagram. Figure (4.3) is a complete phase plane diagram for the
logistic predator-prey model.
By analyzing the phase plane, numerous anomalies can be found. One such anomaly is the
fact that (x2 , y2 ) is a stable spiral instead of a center. Recall that in the normal Lotka-Volterra
model, the steady state with x = α2 /γ2 was a center, implying that the populations of both
species would oscillate indefinitely. In the logistic population model, however, the corresponding
steady state is a stable spiral, which means that eventually the populations of both species would
become constant. In fact, when the prey population grows logistically, as t → ∞, x → αγ22 and
y → αγ11 − γα11γα2 K
2
. In mathematical notation:
α2
lim x(t) = (4.36)
t→∞ γ2
α1 α1 α2
lim y(t) = − (4.37)
t→∞ γ1 γ1 γ2 K
Since K > α2 /γ2 , notice that the steady state (x2 , y2 ) always has a prey population less
than the carrying capacity of the prey. In other words, even though the prey population can
have a maximum population of K without the presence of the predator, when the predator is
present the prey population may no longer support a population of K.
In the regular Lotka-Volterra model, the nontrivial steady state depended exclusively on the
parameters α1 , α2 , γ1 , and γ2 , but not on the populations of either the prey or predator. In
the logistic model, similar behavior is observed, except (x2 , y2 ) depends on K in addition to
the four other parameters. The fact that the stability of the logistic model does not depend
on the population of either species is quite remarkable. No matter how large or how small the
initial populations of both species are, as long as the parameters are the same, eventually the
populations will tend to the same steady state.
Finally, remember that the regular Lotka-Volterra model does not accurately predict the
behavior of a predator-prey system when one or both species are endangered. The same disad-
vantage holds for the logistic model as well. When the prey or the predator population is very
32
Figure 4.2: This is a plot of the nullclines, steady stats, and direction fields on the nullclines for
the logistic predator-prey model. Note that the direction fields are vertical on the f-nullclines
and horizontal on the g-nullclines.
33
Figure 4.3: This is the phase plane diagram for the logistic predator-prey model. Here, α1 =
α2 = γ1 = γ2 = 0.5 and K = 2. Steady states are (0,0), (2,0), and (1,0.5).
34
small (when the solution curves are very close to the x or y-axis),the logistic model predicts
that the population of the species will always rebound to safer levels. The logistic model rarely
predicts that either species will become extinct, even when the solution starts very close to the
origin!
4.5 Conclusion
In this chapter, we made a small modification in the regular Lotka-Volterra model so that
the prey population grew logistically instead of exponentially. A logistic model for population
growth is more realistic than an exponential model because environmental limitations typically
suppress the population growth of species. By creating a phase plane diagram, we discovered
that the small modification created a very sizable difference. In the regular Lotka-Volterra
model, there were only two steady states, but in the logistic model, there were three. More
significantly, the regular Lotka-Volterra model predicted that the populations of both species
will oscillate indefinitely, but the logistic model predicts that the populations of both species
will eventually reach an equilibrium point. This stability does not depend on the populations
of either species. Rather, the stability depends on the parameters α1 , α2 , γ1 , γ2 , and K. In
the phase plane analysis, we assumed that K > α2 /γ2 , so that the steady state (x2 , y2 ) was in
the first quadrant. In one of the problems below, you will discover what will happen if that
study state has no biological significance (i.e. when (x2 , y2 ) is not in the first quadrant and
K < α2 /γ2 ).
4.6 Problems
1. In Problem (1) in the previous chapter, the symbiotic model was introduced. Show that
the first two terms on the right hand side of both equations, α1 x − β1 x2 and α2 y − β2 y 2 ,
model logistic growth by using Equation (4.1) and relating β, α, r, and K. In terms of α
and β, what is the carrying capacity of both species when the other species is not present?
2. In the phase plane analysis for Equations (4.4) and (4.5), we assumed that K > α2 /γ2 .
In this problem, you will consider what happens when K < α2 /γ2 .
(a) If K < α2 /γ2 , what quadrant is (x2 , y2 ) in? Does this steady state have any biological
significance? Why or why not?
(b) Does the stability of the other two steady states change with this assumption on K?
(c) Draw the resulting phase plane diagram assuming K < α2 /γ2 . Graph only the
first quadrant since x and y must be greater or equal to zero. (Hint: Graphing the
nullclines and the direction fields on the nullclines will make it easier to draw solution
curves.)
(d) Explain, in biological terms, what the logistic model predicts when K < α2 /γ2 .
3. A generalized logistic equation is:
v
dP P
= rP 1− (4.38)
dt K
35
Equation (4.38) has the solution:
K
P (t) = 1/v
(1 + Qe−rv(t−t0 ) )
where:
v
K
Q = −1 +
P0
and P0 is the initial population of a species at the initial time t0 .
(a) Suppose t0 = 0, P0 = 10, K = 100, and r = 0.05. Graph P (t) when v = 1, v = 2 and
v = 3 to determine the effect v has on P (t). See Figure (4.1) as a reference. You
may use a calculator or a computer to graph all three functions, but you must hand
in a sketch of the graphs correctly labeled.
(b) If we modify the Lotka-Volterra model for the case where v = 2, we get:
x 2
dx
= α1 x 1 − − γ1 xy
dt K
dy
= −α2 y + γ2 xy
dt
Suppose α1 = α2 = γ1 = γ2 and K = 2.
i. Determine all the nullclines for the system of equations.
ii. Identify the steady states and determine their stability.
iii. Graph the phase plane for the system of equations, considering only the first
quadrant. Include all the nullclines in your phase plane.
iv. In biological terms, what does the model predict?
v. Figure (4.3) is a phase plane diagram for the case where v = 1. How do the two
cases compare?
36
Chapter 5
Populations in Competition
5.1 Introduction
In the previous two chapters, we discussed two predator-prey models, in which the predator
survives by eating the prey. This is one of many interactions that may occur between two
different species. Another type of interaction between two (or more) species is modeled with
the competition model. In the competition model, both species compete with one another
because they utilize the same resources. Examples of resources are food, shelter, and territory.
One species does not eat the other species.
Two different outcomes can occur when two or more species compete for the same resources.
Either a dominant species that is able to exploit resources the most efficiently prevails and the
other species become extinct or there is no dominant species and all species are able to coexist.
The competition model tells us under what circumstances each of these possibilities will occur.
5.2 The Competition Model

Let N1 be the population density of one species, N2 be the population density of another species,
and t be time. Assuming that the population of both species will grow logistically without the
presence of the competing species, we can write:
dN1 K1 − R1
= r1 N1 (5.1)
dt K1
dN2 K2 − R2
= r2 N2 (5.2)
dt K2
where r1 is the rate at which Species 1 approaches its carry capacity K1 , and R1 is the number
of individuals that are using the resources that Species 1 uses. Similarly, r2 is the rate at which
Species 2 approaches its carrying capacity K2 , and R2 is the number of individuals that are using
the resources that Species 2 uses. Of course, without the presence of the competing species,
R1 = N1 and R2 = N2 .
37
However, when the competing species is present, R1 and R2 become:
R1 = N1 + B2 N2 (5.3)
R2 = N2 + B1 N1 (5.4)
where B1 is a parameter that is used to relate how much Species 1 uses the resources that
Species 2 uses. If B1 = 0, then Species 1 does not use the resources of Species 2. If B1 = 1,
then Species 1 uses the same amount of resources per organism as Species 2. If B1 < 1, then
Species 1 uses less resources per organism than Species 2. Finally, if B1 > 1, then Species 1 uses
more resources per organism than Species 2. B2 behaves similarly, except B2 is a parameter
that relates how much Species 2 uses the resources that Species 1 uses. For the competition
model we will assume that B1 > 0 and B2 > 0.
The competition model can be derived by substituting Equation (5.3) into Equation (5.1)
and substituting Equation (5.4) into Equation (5.2). The competition model is:
dN1 K1 − N1 − B2 N2
= r1 N1 (5.5)
dt K1
dN2 K2 − N2 − B1 N1
= r2 N2 (5.6)
dt K2
Notice that the competition model does not make a reference to the resources that the species
share. Instead, it is assumed that the presence of each species leads to the suppression of the
competitor’s population growth. Table (5.1) summarizes all of the parameters in the competition
model. Note that all of the parameters are positive.
38
Table 5.1: Description of Parameters in the Competition Model
r1 Determines how quickly Species 1 approaches its car- 1/time
rying capacity. The greater r1 is, the faster Species 1
approaches its carrying capacity.
r2 Determines how quickly Species 2 approaches its car- 1/time
rying capacity. The greater r2 is, the faster Species 2
approaches its carrying capacity.
K1 The carrying capacity of Species 1. density
K2 The carrying capacity of Species 2. density
B1 Measures how many resources that Species 1 uses per unitless
individual that Species 2 uses. If B1 > 1, then Species
1 uses more resources per individual than Species 2.
If B1 < 1, then Species 1 uses less resources per indi-
vidual than Species 2.
B2 Measures how many resources that Species 2 uses per unitless
individual that Species 1 uses. If B2 > 1, then Species
2 uses more resources per individual than Species 1.
If B2 < 1, then Species 2 uses less resources per indi-
vidual than Species 2.
5.3 Nondimensionalization
We could continue the normal procedure of finding the nullclines, steady states, and their
stability using Equations (5.5) and (5.6), however, computations will become very tedious.
If you are not convinced of this, try to compute the Jacobian for the system of differential
equations! To make the computations easier, we can perform nondimensionalization on the
system of equations. To do this, first we need to separate all of the variables into a unitless part
and a unit carrying part, as such:
Measured Quantity = Scalar Multiple × Unit Carrying Dimension

N1 = N1∗ × N̂1
N2 = N2∗ × N̂2
t = t∗ × τ
Now, we can rewrite Equations (5.5) and (5.6) as such:
!
d(N1∗ N̂1 ) K1 − N1∗ N̂1 − B2 N2∗ N̂2
= r1 (N1∗ N̂1 ) (5.7)
d(t∗ τ ) K1
!
d(N2∗ N̂2 ) K2 − N2∗ N̂2 − B1 N1∗ N̂1
= r2 (N2∗ N̂2 ) (5.8)
d(t∗ τ ) K2
39
To make the left hand side of the equations above unitless, we can divide by N̂ and multiply
by τ :
!
dN1∗ τ K 1 − N ∗
1 1N̂ − B N ∗
2 2 2 N̂
= r1 (N1∗ N̂1 ) (5.9)
dt∗ N̂1 K 1
!
dN2∗ τ K 2 − N ∗
N̂2 − B1 N ∗
N̂1
= r2 (N2∗ N̂2 ) 2 1
(5.10)
dt∗ N̂2 K 2
So far, all that we managed to do is make a mess out of things! To simplify the hideous equations
above:
1
Let N̂1 = K1 , N̂2 = K2 , and τ= (5.11)
r1
Note that the units are equivalent on both sides of all three equations. It would be incorrect
if that was not the case. Also, note that the left hand sides of all three equations contain only
variables, and the right hand sides contain only parameters. We cannot assign one variable to
another variable, but we may assign a variable to a parameter. Substituting Equations (5.11)
into Equations (5.9) and (5.10) yields:
!
dN1∗ B N ∗
2 2 2N̂
= N1∗ 1 − N1∗ − (5.12)
dt∗ K1
!
dN2∗ B N
1 1 1
∗
N̂
= τ r2 N2∗ 1 − N2∗ − (5.13)
dt∗ K2
We can simplify the equations further by making the following substitutions:

B2 B1
Let ρ = τ r2 , γ= N̂2 , and β= N̂1 (5.14)
K1 K2
Notice that the right hand side of the three equations above are unitless, so ρ, β, and γ must
be unitless as well. Equations (5.12) and (5.13) simplify to:
dN1∗
= N1∗ (1 − N1∗ − γN2∗ ) (5.15)
dt∗
dN2∗
∗
= ρN2∗ (1 − N2∗ − βN1∗ ) (5.16)
dt
Finally, we can simplify the system of equations further by letting u = N1∗ and v = N2∗ . This is
an optional step, but by making these substitutions, we no longer have to write the subscripts
or the stars. The final dimensionless competition model is:
u̇ = u(1 − u − γv) (5.17)

v̇ = ρv(1 − v − βu) (5.18)
where u̇ is the derivative of u with respect to t∗ and v̇ is the derivative of v with respect to t∗ .
40
Although it took some effort to produce a dimensionless form of the competition model, we will
now relish the fruits of our labor by breezing through this section. If we did not execute the
nondimensionalization procedure for the competition model, the computations in this section
would be brutal. We begin as always by finding the nullclines of the system.
The u-nullclines are:
u̇ =0 = u(1 − u − γv) (5.19)

u=0 or 0 = 1 − u − γv (5.20)
1−u
v= (5.21)
γ
The v-nullclines are:
v̇ =0 = ρv(1 − v − βu) (5.22)

v=0 or 0 = 1 − v − βu (5.23)
v = 1 − βu (5.24)
Next, we may find the steady states of the system by identifying the intersections of the
u-nullclines with the v-nullclines. Three of them are:
(u1 , v1 ) = (0, 0) (5.25)

(u2 , v2 ) = (0, 1) since u=0 implies that v = 1 − βu = 1 (5.26)
(u3 , v3 ) = (1, 0) since v=0 implies that u = 1 − γv = 1 (5.27)
There is another possible steady state, (u4 , v4 ), which is left as an exercise to find.
The next step is to find the Jacobian of the system at each steady state:

(1 − u − γv) − u −γu
J(u, v) = (5.28)
−ρβv ρ(1 − v − βu) − ρv

1 0
J(0, 0) = (5.29)
0 ρ

−1 −γ
J(1, 0) = (5.30)
0 ρ(1 − β)

1−γ 0
J(0, 1) = (5.31)
−ρβ −ρ
Next, we can find the eigenvalues for each steady state:

For (0,0):
det(J(0, 0) − λI) = (1 − λ)(ρ − λ) = 0 (5.32)

∴ λ1 = 1 > 0 and λ2 = ρ > 0 (5.33)
41
For (1,0):
det(J(1, 0) − λI) = (−1 − λ)(ρ(1 − β) − λ) = 0 (5.34)

∴ λ1 = −1 < 0 and λ2 = ρ(1 − β) (5.35)
For (0,1):
det(J(0, 1) − λI) = (1 − γ − λ)(−ρ − λ) = 0 (5.36)

∴ λ1 = −ρ < 0 and λ2 = 1 − γ (5.37)
Instead of actually calculating the eigenvalues above, we could have alternatively read them
off the diagonal of the three matrices instead since the matrices are triangular. Note that for
the steady state (0,0), λ1 > 0 and λ2 > 0. Therefore (0,0) must be an unstable node. The
stability of the other two steady states depends on the sign of λ2 . The sign of λ2 depends on γ
and β. This leads to four cases:
• Case 1: β < 1 and γ > 1 (0,1) is a stable node, but (1,0) is a saddle
• Case 2: β > 1 and γ < 1 (1,0) is a stable node, but (0,1) is a saddle
• Case 3: β > 1 and γ > 1 (1,0) and (0,1) are stable nodes
• Case 4: β < 1 and γ < 1 (1,0) and (0,1) are saddles

We will now attempt to draw a phase plane for each of the four cases. The first step is to graph
the nullclines for each case, in addition to the direction fields located on the nullclines. All
direction fields on the u-nullclines will be vertical, since du
dt
= 0 on those lines, and all direction
dv
fields on the v-nullclines will be horizontal, since dt = 0 on those lines. Remember that direction
fields on a nullcline may change directions only if they cross a steady state. Figure (5.1) shows
the nullclines, direction fields, and steady states of each of the four cases. Also, notice that
(u4 , v4 ) is only present in Cases 3 and 4. This is because in the other two cases, (u4 , v4 ) is not
in the first quadrant and therefore has no biological significance. Please study the four graphs
before proceeding.
Using the direction fields and the stability of each steady state, it is very easy now to
draw solution curves and complete the phase plane diagrams. Figure (5.2) shows phase plane
diagrams for all four cases in the competition model. For the phase planes, ρ = 1 and values of
1 and 1.5 are used for γ and β, depending on the specified inequalities for each case.
Before we can perform an analysis on the phase planes, first we must convert each steady state
back into the original parameters. Although nondimensionalization has made the mathematical
components of this procedure easier, we must revert all the important points, such as the steady
states, back into its original form so that we can interpret them in a biological context. For
instance, you can do all of this phase plane analysis without reverting the parameters, and tell a
biologist, ”The solutions approach (1,0) when β > 1 and γ < 1!” but this would be meaningless
42
Figure 5.1: These are graphs of the nullclines, direction fields on the nullclines, and steady
states for each of the four cases.
43
Figure 5.2: Above are the phase plane diagrams for each case of the competition model.
44
to the biologist. A mathematician must be able to communicate the results of his or her work
in the proper context.
Let us begin by converting the steady state (1,0) back into its original variables. What the
point (1,0) actually means is u = 1 and v = 0. Recall that we assigned u = N1∗ and v = N2∗ , so
N1∗ = 1 and N2∗ = 0. The original variables were N1 = N1∗ × N̂1 and N2 = N2∗ × N̂2 , but N̂1 = K1
and N̂2 = K2 , so N1 = N1∗ × K1 and N2 = N2∗ × K2 . Finally, since (u, v) = (N1∗ , N2∗ ) = (1, 0),
(N1 , N2 ) = (K1 , 0).
For (0,1) and (0,0), the procedure is the same:
(0, 1) = (u, v) = (N1∗ , N2∗ ) ∴ (N1 , N2 ) = (N1∗ × K1 , N2∗ × K2 ) = (0 × K1 , 1 × K2 ) = (0, K2 )

(0, 0) = (u, v) = (N1∗ , N2∗ ) ∴ (N1 , N2 ) = (N1∗ × K1 , N2∗ × K2 ) = (0 × K1 , 0 × K2 ) = (0, 0)
It is left as an exercise to convert (u4 , v4 ) back into its original form. Table (5.2) list the
steady states in their unitless form and their original form.
Table 5.2: Steady States in their Unitless and Original Forms

Unitless Form Original Form
(0,0) (0,0)
(1,0) (K1 , 0)
(0,1) (0, K2 )
(u4 , v4 ) left as an exercise
We are finally ready to perform a phase plane analysis for each of the four cases.
CASE 1: Of the three steady states, only (N1 , N2 ) = (0, K2 ) is stable. As t → ∞, all
solutions approach (0, K2 ). This means that eventually the species N1 will become extinct and
the species N2 will prevail and attain its carrying capacity K2 . To understand why this occurs,
let us consider the conditions in which Case 1 exist: β < 1 and γ > 1. Of course, from (5.14)
β = K1 B1 /K2 and γ = K2 B2 /K1 , so Case 1 exist under the conditions K1 B1 /K2 < 1 and
K2 B2 /K1 > 1. If we assume for a moment that K1 = K2 , then the inequalities become B1 < 1
and B2 > 1. Therefore, B2 > B1 and species N2 is very aggressive in obtaining the resources
that species N1 uses. It is interesting to note that irregardless of the initial populations of both
species, species N2 will always prevail and species N1 will become extinct so long as the two
inequalities hold. That is to say, the stable steady state does not depend on the population
of either species, but only on their carrying capacities and their aggressiveness in obtaining
resources that the competing species needs.
CASE 2: The only steady state that is stable in this case is (N1 , N2 ) = (K1 , 0). Case 2 is the
opposite of Case 1. As t → ∞, all solutions approach (K1 , 0). This means that eventually the
species N2 will become extinct and the species N1 will prevail and attain its carrying capacity
K1 . In order for Case 2 to exist, β > 1 and γ < 1. In terms of the original parameters,
K1 B1 /K2 > 1 and K2 B2 /K1 < 1. Assuming K1 = K2 , the inequalities reduce to B1 > 1 and
B2 < 1. Therefore, B1 > B2 and species N1 is very aggressive in obtaining the resources that
45
species N1 uses. Notice again that no matter what the initial populations of both species are,
the solutions will approach the stable steady state as long as the two inequalities hold.
CASE 3: In this case, there are two stable steady states, (N1 , N2 ) = (K1 , 0) and (N1 , N2 ) =
(0, K2 ). Notice for Case 3, the inequalities are K1 B1 /K2 > 1 and K2 B2 /K1 > 1. Assuming
K1 = K2 , both B1 and B2 are large, so both species are aggressive competitors. This case is
unique in that the initial populations of both species do effect which of the two stable steady
states a particular solution approaches. By studying the phase plane for Case 3 in Figure (5.2)
it becomes apparent that the graph can be partitioned into two regions separated by a line
passing through the origin and (u4 , v4 ). This line is called a separatrix because it lies exactly
where the behavior of the system changes. In the lower region below the separatrix, notice
that all solution curves approach (K1 , 0). Therefore, if species N1 has a huge initial population
compared to species N2 , then species N1 will prevail, reaching its carrying capacity, and species
N2 will become extinct. On the other hand, in the upper region above the separatrix, all the
solution curves approach (0, K2 ). Thus, if species N2 has a huge initial population compared
to species N1 , then species N2 will prevail, reaching its carrying capacity, and N1 will become
extinct. In Figure (5.2), the separatrix is labeled in green.
CASE 4: In this case, the only stable steady state is (N1 , N2 ) = (u4 , v4 ). In order for Case
4 to exist, K1 B1 /K2 < 1 and K2 B2 /K1 < 1. Assuming K1 = K2 , both B1 and B2 are small, so
both species are not very aggressive competitors. Thus, as t → ∞, all solution curves approach
(u4 , v4 ). In fact, (u4 , v4 ) is a stable node in this case. Although it is left as an exercise to find
(u4 , v4 ), it is apparent by looking at the phase plane for Case 4 in Figure (5.2) that the steady
state is located to the left of (K1 , 0) and below (0, K2 ). This means that both species can
achieve coexistence when the two inequalities above are satisfied, however, neither of them will
be able to attain their carrying capacities because of the presence of their competitor. Moreover,
coexistence is only possible because both species are weak competitors. Notice that in this case,
the solution curves will always approach (u4 , v4 ), regardless of the initial populations of both
species.
5.6 Conclusion
In this chapter, we considered another system that models the interaction of two species. This
time, we considered the case in which two species compete with one another over the same re-
sources. The system of differential equations for this system was developed under the assumption
that without the presence of the competing species, each population will grow logistically. In
order to simply the mathematics in this section, we applied a process called nondimensional-
ization. Performing nondimensionalization simplifies the system of equations by eliminating all
the parameters with units. However, it is important to revert to the original parameters when
interpreting the system in biological terms. By finding the eigenvalues of the steady states
for the system, we discovered that there are four possible cases, each with unique biological
consequences. The four cases depend on the carrying capacities and the aggressiveness of both
species in competing for resources. According to the competition model, if one species is a
very aggressive competitor, then that species will prevail, obtaining its carrying capacity, while
the other species will become extinct. If both species are aggressive competitors, then either
species may prevail, depending on the initial populations of the species. If one species has a
46
significantly larger initial population than the other species, then the species with the larger
initial population will dominate and attain its carrying capacity while the competing species
becomes extinct. Finally, if neither species are very aggressive, then both species may coexist,
but neither of them will be able to attain their carrying capacities.
5.7 Problems
1. In this problem, you will find the steady state (u4 , v4 ) and determine its stability.
(a) Compute the steady state (u4 , v4 ) by finding the intersection of the u-nullcline v =
(1 − u)/γ and the v-nullcline v = 1 − βu.
(b) For Case 3, determine the stability of (u4 , v4 ).
(c) Revert the parameters in (u4 , v4 ) into the original parameters B1 , B2 , K1 and K2 .
2. Perform nondimensionalization on the Lotka-Volterra Equations (3.1) and (3.2) into the
following system:
dv
= v(1 − e)
dt
de
= αe(v − 1)
dt
Clearly define each parameter and variable assignment.
3. In the last three chapters, you have seen various examples of systems of differential equa-
tions that model the interaction between two species. In this problem, please write a
system of three differential equations, one for each species, that model the situation be-
low:
Suppose x is a predator and y and z are both its prey. Species z grows logistically in
the absence of its predator. Species x dies out in the absence of prey. Species y grows
exponentially in the absence of the predator.
47
Chapter 6
Models for Infectious Diseases
6.1 Introduction
Throughout all of recorded history, epidemics have plagued the human population. One of the
earliest mathematical models that was used to predict the course of epidemics was the Kermack
and McKendrick model, also known as the SIR model. In 1926, the Kermack and McKendrick
model was used to model the death rate due to a 1906 plague in Bombay, India. The SIR
model is a system of three nonlinear differential equations. The equations model the number of
individuals in three categories: susceptible, infected and recovered. The susceptible population
are people that have not been infected yet, but may become infected if they are in contact with
the infected. The infected population are the individuals carrying the disease, and can infect
susceptible o. The recovered population are individuals that have become infected but have
cured. In the SIR model, it is assumed that the recovered individuals gain permanent immunity
to the disease. There are many illnesses in which people do acquire such immunity (such as the
chicken pox and the 1906 Bombay plague). However, people do not become immune to some
illnesses (such as the common cold), so the SIR model would not be a good model for such
sicknesses. Figure (6.1) illustrates the progression of an epidemic using the SIR model. Notice
the two proportionality constants, β and υ. β is the rate of infection, the probability that
a susceptible individual will become infected when in ”‘contact”’ with an infected individual.
”‘Contact”’ can mean a handshake with an infected person, being close to a sneezing infected
person, having sexual intercourse with an infected person, etc., depending on how the particular
disease is spread. υ is the rate of recovery, the average proportion of infected individuals that
become recovered per period.
48
Figure 6.1: This diagram shows the progression of an epidemic using the SIR model. The
susceptible individuals become infected at a rate of β individuals per period and the infected
become recovered at the rate of υ individuals per period.
6.2 Models for Infectious Diseases

The SIR model outlined in Figure (6.1) is the following system of differential equations:
dS
= −βSI (6.1)
dt
dI
= βSI − υI (6.2)
dt
dR
= υI (6.3)
dt
Note that in the above system, β is multiplied by S and I, while υ is multiplied by only I.
This is because the rate of transmission from susceptible individuals to the infected depends on
both the susceptible and the infected populations (i.e. the higher the susceptible and infected
populations are, the faster the transmission), but rate of recovery only depends on the number
of infected individuals. In the SIR model, there are no births or deaths. Also note that in the
SIR model, the total population, N , is conserved. That is to say, N = S + I + R. N is constant
because dS/dt + dI/dt + dR/dt = 0 for all time t. Table (6.1) summarizes the parameters used
in the SIR model.
Suppose we wanted to model a disease in which individuals can have temporary, rather than
permanent immunity. This model is appropriately called the SIRS model. In contrast to the
SIR model, in the SIRS model the recovered individuals eventually become susceptible again.
Since the SIRS model is more interesting than the SIR model, the SIRS model will be the focus
of this chapter. Figure (6.2) illustrates the progression of a SIRS epidemic. The SIRS model is
the system of equations below. It is very similar to the SIR model, except we introduce a new
parameter γ to represent the rate of conversion from recovered to susceptible per period.
dS
= −βSI + γR (6.4)
dt
dI
= βSI − υI (6.5)
dt
dR
= υI − γR (6.6)
dt
49
Figure 6.2: This diagram shows the progression of an epidemic using the SIRS model. It is very
similar to the SIRS model, except the recovered individuals may become susceptible again at a
rate of γ individuals per period.
Notice that the SIR model is a special case of the SIRS model when γ = 0. Therefore, all the
conclusions that we make on the SIRS model hold for the SIR model, as long as we assume
that γ = 0. Also notice that the total population is again constant, so that N = S + I + R.
Table (6.1) summarizes the parameters in the SIRS model.
Table 6.1: Description of Parameters in the SIR and SIRS Models

β The rate of infection. The likelihood that a suscep- 1/time
tible individual will become infected when contacted
with an infected individual.
υ The rate of recovery. The proportion of infected in- 1/time
dividuals that recover per one period of time.
γ (SIRS only) The rate of conversion from recovered individuals to 1/time
susceptible individuals.
6.3 Steady Sates for the SIRS Model and their Stability
Even though we have three equations instead of two, it is still possible to find the steady states
using the same method we have been using (i.e. equating all differential equations to zero and
finding their intersection). Equating Equations (6.4)- (6.5) yields:
βSI
R= (6.7)
γ
υI = βSI (6.8)
υI
R= (6.9)
γ
50
Equation (6.8) yields I = 0 or S = υ/β. If I = 0, then Equation (6.9) implies that R = 0. Also
since N = S + I + R, I = 0, and R = 0, then S = N . Thus, one steady state for the SIRS
model is:
(S1 , I1 , R1 ) = (N, 0, 0) (6.10)
If S = υ/β, by Equation (6.9):

υ υI
N =S+I +R= +I +
β γ
γ(βN − υ) υI υ(βN − υ)
∴I= and R= =
β(υ + γ) γ β(υ − γ)
The second steady state for the SIRS model is:

υ γ(βN − υ) υ(βN − υ)
(S2 , I2 , R2 ) = , , (6.11)
β β(υ + γ) β(υ − γ)
The steady state (S1 , I1 , R1 ) means that the entire population is healthy, although everyone is
susceptible. No one in the population is infected and the disease is eradicated. The steady
state (S2 , I2 , R2 ) means that the population is divided into a constant number of susceptible,
infected, and recovered individuals. There is an epidemic and unless the environment improves
to decrease β or to increase υ, the epidemic will remain at that constant level. Note that in
order for (S2 , I2 , R2 ) to have biological significance, S2 , I2 and R2 must be all positive since there
cannot be a negative amount of people. In order for I2 to be positive, N > υ/β. Thus, there is
the following threshold effect:
υ
There will be an epidemic in the population if and only if N >
β
Notice that in order for there to be an epidemic, the total population N must be large enough
so that the inequality above is true. In other words, a an epidemic will not occur if the total
population is very small.
Now, in order to draw two-dimensional phase planes, we must reduce the system into two
equations. We may use the fact that N = S + I + R to accomplish this. By substituting
R = N − S − I into Equations (6.4) and (6.5), we can eliminate R and get:
dS
= −βSI + γ(N − S − I) := f (S, I) (6.12)
dt
dI
= βSI − υI := g(S, I) (6.13)
dt
The nullclines can be easily determined by equating f (S, I) = 0 and g(S, I) = 0, which yield:
γ(N − S)
f-nullcline: βSI = γ(N − S − I) → I = (6.14)
βS + γ
g-nullclines: I=0 and S = υ/β (6.15)
51
We can find the steady states by finding the intersection of the f-nullclines with the g-nullclines,
but an easier method is to read off the values of S and I from Steady States (6.10) and (6.11).
The steady sates are:
(S1 , I1 ) = (N, 0) (6.16)

υ γ(βN − υ)
(S2 , I2 ) = , (6.17)
β β(υ + γ)
To determine the stability of these two steady states, first we need to calculate the Jacobian.
The first column is the partial derivatives of S with respect to t of both equations, and the
second column is the partial derivatives of I with respect to t of both equations.

−βI − γ −βS − γ
J(S, I) = (6.18)
βI βS − υ

−γ −βN − γ
J(S1 , I1 ) = (6.19)
0 βN − υ
 βγ(βN −υ) 
− βυ+γ − γ −υ − γ
J(S2 , I2 ) =   (6.20)
βγ(βN −υ)
βυ+γ
0
Now we can determine the stability of each steady state. For (S1 , I1 ), notice that J(S1 , I1 ) is
an upper triangular matrix, so the eigenvalues are simply the elements on the diagonal, namely
λ1 = −γ and λ2 = βN − υ. Since we are assuming that N > υ/β and since all the parameters
must be positive, it follows that λ2 > 0 > λ1 . Thus, (S1 , I1 ) is a saddle.
For (S2 , I2 ):
βγ(βN − υ)
tr(J(S2 , I2 )) = − −γ <0 (6.21)
βυ + γ
det(J(S2 , I2 )) = βI2 (υ + γ) > 0 (6.22)
Since tr(J(S2 , I2 )) is always negative and det(J(S2 , I2 )) is always positive, then (S2 , I2 ) is always
stable when it exist (i.e. when N > υ/β).

By drawing the nullclines with direction fields, it is easy to determine if (S2 , I2 ) is a stable
node or a stable spiral. Figure (6.3) gives clear evidence that (S2 , I2 ) is a stable spiral, since
solution curves must be horizontal on the nullcline S = υ/β. If (S2 , I2 ) were a stable node, then
the solution curves have to be vertical on the nullcline. It is very easy to draw a phase plane
using Figure (6.3). The phase plane is shown in Figure (6.4). As always, we are only concerned
with the first quadrant since the number of susceptibles and the number of infected cannot be
negative.
Although the phase plane does not have an axis for R, it is still possible to determine the
number of recovered individuals since R = N − S − I. In Figure (6.4), for example, N = 2 so
on the steady state (S2 , I2 ) = (1, 0.5), R = 2 − 1 − 0.5 = 0.5. (Notice that R is a fraction. This
52
Figure 6.3: This is a graph of the nullclines of the SIRS model. Horizontal and vertical direction
fields on the nullclines are also show. Note that (S2 , I2 ) must be a spiral, not a node, because
the nullcline S = υ/β has only horizontal direction fields.
53
Figure 6.4: This is a phase plane diagram for the SIRS model, where β = γ = υ = 0.25 and
N = 2. The f-nullcline is in pink and the g-nullclines are in orange. Steady states are (2,0) and
(1, 0.5). Note that N υ/β > 1 so both steady states exist and there is an epidemic.
54
is okay if we have scaled the variables so that every unit equals, say, 1000 individuals. So, the
total population would be N = 2 ∗ 1000 = 2000, R = 0.5 ∗ 1000 = 500, etc. Scaling variables is
very common in mathematical models involving large numbers.)
Although the phase plane for the SIRS model is fairly simple, a phase plane analysis produces
a few surprising results. One odd result is the fact that when there is an epidemic (N υbeta¿1),
then as time gets increasingly large, the number of susceptible and infected individuals stabilizes
into a constant value. In particular, the number of susceptible and infected individuals is
determined by (S2 , I2 ). This is counter-intuitive for two reasons. For one, since individuals go
from S → I → R, and then back to S, it is remarkable that the population in each group
becomes constant as time becomes large. Also, it should be expected that as time becomes
large that the epidemic would end (i.e that I = 0). However, since the parameters β, γ, υ, and
N are assumed to be constant in the model, the epidemic never ends. Rather, the epidemic
stabilizes, where the number of infected becomes I2 . If, on the other hand, the infection rate
(β) or the rate of conversion from R to S (γ) decreases or the recovery rate (υ) increases, then
the epidemic may end. In other words, the only way to stop an epidemic without a vaccination
is to improve environmental health conditions.
Another unexpected results is that the number of susceptible, infected, and recovered indi-
viduals depends only on υ, β, γ, and the total population N , no matter how many susceptible,
infected, or recovered individuals there were at the initial time (as long as I 6= 0). Even if the
number of infected individuals equaled the total population, the population will still stabilize
to the steady state (S2 , I2 )!
6.5 Conclusion
There are many mathematical models that illustrate the progression of epidemics. One such
model is the SIRS model, in which a population is categorized under three groups: susceptible,
infected, and recovered. In the SIRS model, the infected infects the susceptible individuals
and the infected eventually recover. However, recovery is not permanent and the recovered
individuals may become susceptible again. Also, the SIRS model does not assume that any
births or deaths occur in the population, so that the total population is always equal to the
number of susceptible, infected, and recovered individuals. Whether or not an epidemic occurs
depends on the quantity N β/υ. If N β/υ < 1, then there is no epidemic, but if N β/υ > 1, then
there is an epidemic. By performing a phase plane analysis on the later case, it is apparent
that the only way to stop an epidemic is increase υ, decrease β, or decrease γ. If all of these
parameters remain constant, then the number of infected individuals will stabilize to I2 .
6.6 Problems
1. If N β/υ < 1, then I2 becomes negative and so the steady state (S2 , I2 ) has no biological
significance. However, when N β/υ > 1 we know that (S2 , I2 ) is stable and all solution
curves approach it as time approaches infinity. So, why is it that when N β/υ < 1, the
solution curves do not approach (S2 , I2 )? Is (S2 , I2 ) still stable? Also, is (S1 , I1 ) stable if
55
N β/υ < 1? If there is no epidemic, as time approaches infinity where do the solutions go
to?
2. For the SIR model in Equations (6.1)to (6.3), do the following:
(a) Eliminate R using the fact that N = S + I + R.
dS dI
(b) Find the nullclines for dt
and dt
.
(c) Identify the steady states for the SIR model.
(d) Determine the biological significance of each steady state. What is the threshold
effect for the SIR model?
(e) Determine if each steady state is stable or unstable assuming there is an epidemic.
What happens as time approaches infinity?
(f) Determine if each steady state is stable or unstable assuming that there is no epi-
demic. What happens as time approaches infinity?
3. Suppose the SIRS model was modified so that the population gave birth to susceptibles
at a rate δ and died at a rate of δ. Figure (6.5) illustrates the progression of this model.
(Assuming that the birth rate is equal to the death rate simplifies the problem considerably,
since N is still constant.)
Figure 6.5: This is a diagram that illustrates the progression of an epidemic using the SIRS
model with births and deaths. The birth rate and the death rate of the total population, δ, are
assumed to be equal.
(a) Develop a system of equations that represents the SIRS model with births and deaths.
(b) Find the steady states for this model.
(c) Determine the threshold effect for the model.
(d) In the case where there is an epidemic, determine if each steady state is stable or
unstable.
(e) Compare the SIRS model without births and deaths with the SIRS model with births
and deaths.
56
Chapter 7
The Chemostat Model
7.1 Introduction
One of the most basic elementary uses of ordinary differential equations in real world modeling
is the Chemostat Model. A chemostat is simply an instrument used to culture bacteria in a
controlled environment. The most useful feature of the chemostat is that the rate of the intake
of the nutrients as well as the outflow of the reservoir can be modified in such a way to change
the culture size of the bacteria inside the chemostat.
Figure 7.1: The Chemostat consists of two connected reservoirs, one which supplies stock nu-
trient and another for the bacteria to grow in.
The chemostat’s name is derived from the idea that the chemical environment in which the
bacteria are being grown is in static. When designing a chemostat system, there are two very
57
important factors that must be considered. The first is that the inflow of nutrients cannot be
too great because too much of any one nutrient may cause the culture to be washed out. The
second is that there must always be enough available nutrients for the culture to grow at a
normal rate. Other factors that are important to when designing the chemostat model are what
type of nutrition to use, the overall flow rate for the culture and the volume of volume of the
growing area. However, these are more so concerns of the biologist and less so concerns of the
mathematician.
The chemostat is used today in many real world experiments regarding studies that involve
bacterial strains. Often times, it is very important for an experimenter to have a living culture
on hand when studying anything that may involve bacteria growth. With a chemostat, it makes
it easier to have bacteria of the same attributes to use in multiple trials to better the quality of
data obtained in an experiment.
7.2 The Model

Why do mathematicians study the chemostat’s mathematical properties? The chemostat lends
itself to two first order differential equations that accurately describe the rate of change for
both the amount of bacteria in the chemostat as well as the rate of change of nutrients in the
chemostat. These two equations are derived after analyzing the chemostat using differential
equations
For this model, assume that there is stock nutrient being pumped into the bacterial culture
chamber at a rate equal to that of the outflow of the bacterial chamber. With this assumption
made, it can be inferred that the volume V is constant. With a constant volume, the reservoir
will never overflow nor will it ever completely drain.
Now to describe this model mathematically, equations must be formed in such a way to
accurately portray a rate of change of both the amount of bacteria and the amount of nutrient
in the chemostat. But before one can make equations, variables must be defined. Below are
the key parameters that these equations will be using. One of the variables that has yet to
be discussed is α. Consider α to be the number of nutrients that are used in producing one
additional unit of bacteria. In other words, each time a new bacteria is produced, α units of
nutrients will be consumed by the bacteria. We will refer to 1/α as the yield which fits well into
the model. This indeed makes α a unitless parameter.
We are almost ready to begin deriving equations. All we must do now is state one assumption.
When considering this model it must be understood that all the nutrients are very well stirred
within both reservoirs. In other words, all sub-volumes of the reservoir can be described with
the same concentration of bacteria and nutrients. The bottom of the tank and the top of the
tank are equal in these regards as if this were not the case spatial variations would have to be
part of this model, but in this approach the only independent variable is time.
7.3 The Equations

To start the modeling process, we look to write an equation describing the rate of change of the
bacteria with respect to time dB
dt
. This rate of change can be expressed in terms of a difference
58
Table 7.1: Chemostat Parameters
Parameter Explanation Units
Mass
N Nutrient Concentration in Chemostat Volume
Mass
N0 Initial Nutrient Content in Reservoir Volume
Number
B Population Density of Bacteria Volume
1
Y = α
Yield Constant Unitless
V Volume of Chemostat Volume
Volume
F Flow Rate Time
between the reproduction and outflow of the bacteria in the chemostat growth chamber.
dB
= KB − F B (7.1)
dt
Growth Rate of Bacteria = Reproduction - Outflow
Here, K is the Reproduction Rate and F is the flow rate of chemostat. To further develop
this model another assumption must be made. One can assume that there is only one growth
limiting nutrient being supplied to the bacteria inside the chemostat and that the growth of
the bacteria solely depends on the availability of this nutrient. While it is more common to
have bacteria growing on a variety of nutrients, it is true that usually there is one dominating
nutrient and the rest of these are negligible.
Now that a preliminary equation has been written for the rate of change in the number of
bacteria, we must write an equation for the rate of change in N , the amount of nutrients in the
bacterial growth chamber.
Now we will write a differential equation for the rate of change with respect to N , the rate
of change of the nutrient concentration. Here we will represent α as mentioned in the previous
table as the amount of nutrient used in one population increase.
dN
= −αK(N )B − F N + F N0 (7.2)
dt
Growth Rate of Nutrient= - nutrients used during growth - nutrients lost in outflow + nutrients
gained from intake
While Equations (7.1) and (7.2) are a good start in deriving a model, they are not exactly
correct. To see why they are incorrect we must analyze them using dimensional analysis; a
method which looks to see what the units are on each side of the equation. For dimensional
analysis to be verified, each side of the equation must match. For instance if the left side of
the equation’s unit was in pounds and the right side was in feet, then the equation would fail
dimensional analysis.
59
dB
Dimensional Analysis of dt
Here we have the dimensional analysis of the previous differential equation.

dB
= KB − F B
dt
number 1 number volume number
= − (7.3)
volume × time time volume time volume
Here after performing dimensional analysis it appears as though the last term on the right
side of the equation is not of the same dimension as the other two terms. To solve this problem
F B must be divided by a volume. The new right most term in the equation will be F B/V This
is indeed the density of bacteria that leaves per unit time.
By applying this same analysis to the above equation, it is evident that F N and F N0 also
need to be divided by volume. After this adjustment is made, two new equations can be written.
dB FB
= K(N )B −
dt V
dN FN F N0
= −αK(N )B − +
dt V V
Now that the equations are dimensionally sound, they are ready to be used in modeling.
However, passing dimensional analysis does not necessarily mean that the model is correct on
its physical principles.
Often times in mathematical modeling equations are written in ways that model a physical
property. In this case a great way to start is to use the principle that in a closed system mass
is conserved.
Given that

dB Kmax N FB
= B− (7.4)
dt Kn + N V

dN Kmax N FN F N0
= −α B− + (7.5)
dt Kn + N V V
To develop into the next form of our equations we must use a tool that is known as a mass
balance. Using the scientific principle that in a closed system mass is conserved you can balance
the mass of the nutrients and the number of bacteria in the chamber by writing the equation
in the following manner:
d(BV )
= K(N )BV − F B (7.6)
dt
d(N V )
= −αK(N )BV − F N + F N0 (7.7)
dt
Dividing each equation on both sides by V yields the previous two equations.
60
For a more realistic model of the chemostat, it would be wise to consider the idea that
bacterial growth rates are directly related to the nutrient availability. Per say, there is a threshold
where if there is too much nutrient available to the bacteria, that adding more will not give an
increase in bacterial growth. We call this concept a saturating dependence.
A formula that uses this concept is known as the Michaelis-Menten kinetics. This mechanism
approaches Kmax as an upper bound and KN represents 1/2 of Kmax . K(N ) depicts the rate of
nutrient consumption.
Kmax N
K(N ) = (7.8)
Kn + N
Figure 7.2: Bacterial growth rate and nutrient consumption K(N) modeled by a saturating
function
Now the equations can be modeled in the following more accurate fashion

dB Kmax N FB
= B− (7.9)
dt Kn + N V

dN Kmax N FN F N0
= −α B− + (7.10)
dt Kn + N V V
7.4 Non-Dimensionalization
Every parameter with a unit in a mathematical model can be thought of as a product of two
values. The first is the actual numeric value of the parameter and the second being the unit
61
carrying value of the measurement. For example B the number of bacteria can be thought of
as B ∗ B̂ where B̂ is the unit carrier depicting bacteria amounts and B ∗ is the scalar multiple.
Using this concept it is often possible to reduce the number of variables in a model which will
in turn make the model easier to study, We will redefine our current variables as the following
measured quantity = scalar multiple × unit carrying dimensions

B = B ∗ × B̂
N = N ∗ × N̂
t = t∗ × τ
To begin this simplification of the model, we must first rewrite the model in forms of the
variables last presented in Equations (7.9) and (7.10) only this time these newly defined dimen-
sionless variables are also introduced.
!
d(B ∗ B̂) Kmax N ∗ N̂ F
= B ∗ B̂ − (B ∗ B̂) (7.11)
d(tτ ) Kn + N ∗ N̂ V
!
d(N ∗ N̂ ) Kmax N ∗ N̂ F N ∗ N̂ F N0
= −α B ∗ B̂ − + (7.12)
d(tτ ) Kn + N ∗ N̂ V V
Next, multiply each side of the equation by τ and divide by B̂ and N̂ , respectively. After-
wards, group the constant terms, we have:
!
dB ∗ N∗ τF ∗
= τ Kmax B∗ − B (7.13)
dt∗ Kn /N̂ + N∗ V
! !
dN ∗ −ατ Kmax B̂ N∗ τ F ∗ τ F N0
= B∗ − N + (7.14)
dt∗ N̂ Kn /N̂ + N∗ V V N̂
Now the difficult part of the process begins. To simplify the model from here one must
select values for the unit carrying variables B̂, τ , and N̂ which do not have any value assigned
to them. Choosing the following values will reduce the equation to fewer parameters.
Let
V Kn
τ= , N̂ = Kn , B̂ = ,
F ατ Kmax
Once substituted in, the equations can be written in a new form! The asterisks have been
dropped in order to more clearly show the equations.

dB N
= α1 B−B (7.15)
dt 1+N

dN N
= − B − N + α2 (7.16)
dt 1+N
62
Where:
V Kmax
α1 = (τ Kmax ) = , (7.17)
F
τ F N0 N0
α2 = = . (7.18)
V N̂ Kn
Now the two equations above have two dimensionless parameters α1 and α2 instead of the
original six parameters. (Kn , Kmax , F, V, N0 and α) No matter what substitution is made, there
will always be at least two remaining parameters. Therefore, it can then be said that the model
has two degrees of freedom.
The idea behind reducing the original equations into a simpler form gives a cleaner look of
how the model works. Now when describing a certain instant of the chemostat model, one must
only provide the α values for the equation as those are the only parameters needed to describe
the model. Now that the model has been simplified and reduced, it becomes easier to study
and more importantly; easier to solve. The first part of the analysis here is to look at the model
and see if any steady-state solutions exist.
7.5 Steady State Solutions

A steady state in a mathematical model is a representation of the system where the variables
do not appear to go under any change and are therefore steady. We look for a steady state
solution because we are interested in finding a continuous cycle of bacterial growth inside the
chemostat. Thus time is not a variable of steady state equations and derivatives with respect
to time are equal to zero.
dB
= 0
dt
dN
= 0
dt
From this we may take equations and set them equal to 0. To denote steady-state solution
variables we use the symbols B and N . Below, F is a function of both of these two steady state
variables.

N
F (B, N ) = α1 B−B =0 (7.19)
1+N

N
G(B, N ) = − B − N + α2 = 0 (7.20)
1+N
We are then given two algebraic equations that are readily solvable.
⇒ B = 0 (7.21)
N 1
or = (7.22)
1+N α1
63
Further simplification will show that N = 1/(α1 − 1). When B = 0 it is found that N = α2 .
Otherwise, if B 6= 0 , then G(B, N ) = 0 yields the equation

N
B = (α2 − N ). (7.23)
1+N
Therefore,
1+N
B= (α2 − N ) = α1 (α2 − N ) (7.24)
N
In summary, there are two steady state solutions:

1 1
(B1 , N1 ) = α1 α2 − , (7.25)
α1 − 1 α1 − 1
(B2 , N2 ) = (0, α2 ) (7.26)
The first steady state exists if and only if α1 > 1 and α2 > α11−1 . The second steady state
always exists. We shall show that if the first steady state exists, then it must be stable and
the second steady state is unstable. On the other hand, if the first steady state does not exists,
then the second steady state is stable reflecting the situation that there is no bacteria in the
tank.
7.6 Phase Plane Analysis and Stability

After simplifying the chemostat model we arrived with a system of equations with only 2
parameters, α1 and α2 :

dB N
= α1 B−B
dt 1+N

dN N
= − B − N + α2
dt 1+N
When we found the steady states, we found two types of solutions. The first type of steady
state is trivial and results in all of the bacteria dying! This situation is not interesting to us nor
the researcher. The other steady state relates to a balance of both nutrients and bacteria. To
apply phase phase plane analysis we will focus only on positive values of B and N which will
constrain our direction field and phase plane diagram to only the first quadrant of the B − N
plane.
Step 1: Drawing the Nullclines
A nullcline is a curve which represents where a derivative in a system of ODEs equals 0.
In this case we will focus on where Ḃ and Ṅ equal 0. The nullclines intersection points relate
directly to the steady states and allow us to give shape to our phase plane and direction field.
Ḃ = 0 represents all the points where:
64

N
α1 B−B =0
1+N
Here all of the B = 0 or it is implied that
1
N=
α1 − 1
The other nullcline relates to when Ṅ = 0 where

N
− B − N + α2 = 0
1+N
Solving for N we find that
1+N
B = (α2 − N )
N
To plot this line use the properties that:
1. The curve passes through (0, α2 )
2. The curve is decreasing as a function of B
3. The curve has an asymptote N = 0 as B → +∞
If α1 > 1 and α2 > α11−1 , then the two nullclines intersect at the interior equilibrium (B1 , N1 ).
Otherwise, there is no interior equilibrium point.
The Jacobian matrix at the first steady state (B̄1 , N̄1 ) is:

0 α1 A
J1 =
−1/α1 −(A + 1)
2
where A = N̄1 /(1 + N̄1 ) .
The characteristic equation of this matrix is
det (J1 − λI) = 0 (7.27)
Our characteristic polynomial is:
λ2 + (A + 1)λ + A = 0 (7.28)
This factors into:
(λ + 1)(λ + A) = 0 (7.29)
which yields the eigenvalues:
λ1 = −A, λ2 = −1 (7.30)
Since both of the eigenvalues are real and negative, the steady state is a stable node when it
exists. For the second steady state (B̄2 , N̄2 ), the Jacobian matrix is:

α1 D − 1 0
J2 =
−D −1
where D = α2 /(1 + α2 ). Since this is a lower triangular matrix, the eigenvalues are
λ1 = α1 D − 1, λ2 = −1 (7.31)
α2
The steady state will be a saddle point whenever D > α11 , which is the same as 1+α2
> 1
α1
.
Rearranging, this is the same as α2 > α11−1 assumming that α1 > 1.
65
7.7 Phase Plane Analysis with Direction Fields
Pictured below is the direction field for the Chemostat Model. The direction field can be made
by hand using the methods described above (Nullclines, Steady States, flow etc.). Here the
B-nullclines are colored with blue and the N -nullcline is colored with red. Using the direction
Figure 7.3: Direction Field for the Chemostat Model

.
field, and assigning specific values to parameters to α1 and α2 , one can plot a phase plane
diagram. It is important to remember that when choosing your parameter values that they
must meet the following constraints.
1
α2 >
α1 − 1
α1 > 1
The phase plane diagram for the Chemostat model was plotted using pplane8 which was
downloaded from Professor John Polking’s (Rice University) website.
66
Figure 7.4: Phase Plane for the Chemostat Model with α1 = 1.5 and α2 = 3
.
7.8 Related Questions

1. What is the real world significance of α1 and α2 ? Using dimensional analysis from above
discuss what their units are and how they change when different variables undergo change.
2. In phase-plane analysis, which part of the plot is most relevant to the experimenter? Why?
3. Reduce the original chemostat model to two parameters by naming your variables some
other combination of constants. Is there more than one way to arrive at two degrees of
freedom? Describe your method.
4. In the model discussed, the assumption is made that:

1
< α2
α1 − 1
Model the chemostat and plot the phase plane when this parameter is not met.
67
Chapter 8
Chemical Kinetics
8.1 Introduction
In this chapter, we will turn our attention to a different branch of science - chemistry. More
specifically, we will investigate a branch of chemistry called chemical kinetics. Chemical kinetics
deals with chemical reactions between molecules and atoms and how rapidly reactions occur. For
example, consider one of the most well-known chemical reactions of all, in which hydrogen(H2 )
and oxygen(O2 ) combine to form water(H2 O).
2H2 (g) + O2 (g) → 2H2 O(l)
In this reaction, hydrogen and oxygen are called the reactants because they react to produce
a new substance, in this case water molecules. Water is called the product because it is what is
created from the reaction. The g in parenthesis means that the hydrogen and oxygen molecules
are gases, while the l in parenthesis means that the water molecules are liquid. Hence, it
is not necessary for the reactants or products to be in the same state. Chemical reactions
are stoichiometric, meaning that chemical equations can be written exclusively with natural
numbers. The subscripts in the chemical equation are the number of a specific atom is in a
molecule. So, for example, the molecule H2 contains two hydrogen atoms. The numbers in front
of each symbol represents the quantity of reactants and products that the reaction requires. So,
in order to produce two water molecules, two hydrogen molecules and one oxygen molecule is
required. The chemical reaction above is balanced, meaning that both sides of the arrow contain
an equal number of each atom. On both sides, there are 4 hydrogen atoms and 2 oxygen atoms.
In this chapter, we will consider the rate of reaction for chemical reactions like the one above.
8.2 General Chemical Equations and the Reaction Law

Sometimes, there may be more than one reaction that occurs among the reactants and prod-
ucts and hence there may be more than one chemical equation. Suppose there are i chemical
equations. Then, the general ith chemical equation is:
68
N
X kif N
X
th
i reaction : Vjif Xj Vjir Xj i = 1, 2, ...P (8.1)
kir
j=1 j=1
In Equation (8.1) and in the equations that follow, N is the number of chemicals involved,
P is the number of chemical of chemical equations, and i is the number of chemical equations.
Xj is the j th chemical in the ith equation. Notice that Equation (8.1) has arrows going in both
directions. This means that the reaction is reversible. In this case, the forward reaction is
symbolized by →, while the reverse reaction is symbolized by ←. Vjif is the quantity of Xj
required for the ith forward reaction, while Vjir is the quantity of Xj required for the ith reverse
reaction. As we will see in an example later, V f and V r can be written as N × P matrices. kif
is the rate constant for the ith forward reaction and kir is the rate constant for the ith reverse
reaction.
Before we introduce the equation for the rate of reaction for Xj , let us first define the
quantity (or quantities) qi as follows:
N N
Y f Y r
qi = kif [Xj ] Vji
− kir [Xj ]Vji i = 1, 2, ...P (8.2)
j=1 j=1
In Equation (8.2), [Xj ] is the concentration of chemical Xj . The units of [Xj ] is moles
per liter. A mole is defined to be the amount of a substance that contains as many atoms
or molecules as there are atoms in 12 grams of carbon-12. (Carbon-12 has 6 protons and 6
neutrons.) This number happens to be 6.022 × 1023 . Finally, the equation for the rate of
reaction of Xj is:
P
d[Xj ] X r
Rj := = (Vji − Vjif )qi j = 1, 2, ...N (8.3)
dt i=1
For Equation (8.3), P is the number of chemical equations. Note that a reversible chemical
reaction is considered as one, not two, equations. Equation (8.3) is called the reaction law.
8.3 Rate Law for the Formation of Water

Consider the chemical equation from the Introduction:
1 kf
2H2 + O2 → 2H2 O (8.4)
We will assume that the water produced is not electrified, so that the reverse reaction of
turning water back into hydrogen and oxygen is not possible. Thus, we will assume that the
rate constant for the reverse reaction is zero (i.e. k1r = 0). To calculate the rate law for the
formation of water, we begin by finding the values of P , N , X1 , X2 , and X3 , as shown below:
69
P = 1, N = 3, X1 = H2 , X2 = O2 , X3 = H2 O (8.5)
Then we find V f and V r from Equation (8.4):
V11f = 2 V21f = 1 V31f = 0 (8.6)

V11r = 0 V21r = 0 V31r = 2 (8.7)
There are 2 hydrogen molecules and 1 oxygen molecule on the left side of the Equation (8.4),
so V11f = 2 and V21f = 1. On the right, all we have are two water molecules, so V31r = 2. All the
other Vji s are zero.
V f and V r can be represented in matrix notation as:
   
2 0
V f = 1 V r = 0 (8.8)
0 2
We now have enough information to calculate q1 . Recall that k1r = 0.
q1 = k1f [H2 ]2 [O2 ]1 [H2 O]0 − 0[H2 ]0 [O2 ]0 [H2 O]2 (8.9)
= k1f [H2 ]2 [O2 ] (8.10)
Finally, we can calculate the rate law for each of the chemicals in the chemical equation. We
may simplify notation slightly by letting k = k1f .
d[H2 ]
= R1 = (0 − 2)q1 = −2k[H2 ]2 [O2 ] (8.11)
dt
d[O2 ]
= R2 = (0 − 1)q1 = −k[H2 ]2 [O2 ] (8.12)
dt
d[H2 O]
= R3 = (2 − 0)q1 = 2k[H2 ]2 [O2 ] (8.13)
dt
As one may expect, d[H dt
2]
and d[O
dt
2]
are negative because hydrogen and oxygen are being
d[H2 O]
consumed and dt is positive because water is being produced. Also, notice that the rate
laws for each chemical are neatly related to each other:
1 d[H2 ] d[O2 ] 1 d[H2 O]

− =− = (8.14)
2 dt dt 2 dt
Equations (8.14) allow us to immediately determine the rate of reaction of the other chemi-
cals once the rate of reaction of one chemical is known. Also, an immediate result from Equa-
tions (8.14) is that d[H 2 O]
d[O2 ]
, d[H 2 O]
d[H2 ]
d[O2 ]
, and d[H2]
are constant. Thus the rate of change of one
70
concentration of a chemical with respect to the concentration of another chemical is constant.
This should not be a surprising result since the reaction requires certain amounts of reactants
to produce a certain amount of product. Also, because of this relationship, we only need to
analyze one differential equation (we arbitrarily choose d[Hdt2 O] ), and the analysis of the other
two equations immediately follow.
A natural question to ask, then, is whether or not it is possible to find the concentration of
each chemical at any time t. In other words, is it possible to explicitly solve for [H2 ], [O2 ], and
[H2 O] using Equations (8.11)-(8.13)? To answer this question, we will assume that the initial
concentration of the product, water, is zero. That is, [H2 O]o = 0. Let [H2 ]o and [O2 ]o be the
initial concentrations of hydrogen and oxygen, respectively. Then, using Equation (8.4), we can
establish the following relationships:
[H2 ] = [H2 ]o − [H2 O] (8.15)

1
[O2 ] = [O2 ]o − [H2 O] (8.16)
2
Equation (8.15) comes from the fact that two moles of water requires two mole of hydrogen,
so that one mole of water takes up one mole of hydrogen. Equation (8.16) comes from the fact
that two moles of water requires one mole of oxygen, so that one mole of water takes up one
half mole of oxygen.
Using Equations (8.15) and (8.16), we can rewrite Equation (8.13) as:

d[H2 O] 2 1
= 2k ([H2 ]o − [H2 O]) [O2 ]o − [H2 O] (8.17)
dt 2
What is significant about Equation (8.17) is that we now have a differential equation for
[H2 O] in terms of only the concentration of water and constants. If we are lucky, it may be
possible to explicitly calculate [H2 O] by separation of variables and by using partial fractions.
However, since [H2 ] = ([H2 ]o − [H2 O]) is squared, an explicit formula cannot be derived. Never-
theless, this was not a fruitless approach, because now we have a convenient differential equation
to graph d[Hdt2 O] .
Before we proceed with the graph, however, we may wish to simplify the notation used
in Equation (8.17). To accomplish this, let [H2 O] = u, C1 = [H2 ]o , and C2 = [O2 ]o . Then,
Equation (8.17) becomes:

2 1
u̇ = 2k(C1 − u) C2 − u (8.18)
2
The units for all the constants and variables should be identified. The concentrations [H2 ],
[O2 ], [H2 ]o , and [O2 ]o all have units mol
L
. That makes d[H
dt
2]
and d[O
dt
2] mol
have units L·sec . Therefore,
L2
k must have units mol2 ·sec . Note that the units of k depend on the specific differential equations,
while the units of everything else do not.
71
d[H2 O]
8.4 The Graph of dt
8.4.1 Case 1: [H2 ]o > 2[O2 ]o
To create a graph of the differential equation (8.17), let us consider a few examples. First,
2
suppose that C1 = 5 mol L
, C2 = 2 mol
L
, and k = 0.25 molL2 ·sec . We will consider the case in which
C1 > 2C2 first, then we will consider the case in which C1 < 2C2 . We can sketch a graph of
Equation (8.18), with u on the x-axis and u̇ on the y-axis. Do do this, we can use the tools we
know from algebra and calculus. It is easy to determine that u(t) ˙ intersects with the x-axis at
the points (C1 , 0) and (2C2 , 0) and that it intersects with the y-axis at the point (0, 2kC22 C2 ).
˙ is positive if u < 2C2 and negative elsewhere.
It can also be verified that u(t)
˙ Figure (8.1) is the desired graph. The
We can use this information to plot the graph of u(t).
arrows on the x-axis indicate the change in u as time increases. When u̇ > 0, u increases, so the
arrows point to the right. Similarly, when u̇ < 0, u decreases, so the arrows point to the left.
Figure 8.1: This is the graph of d[Hdt2 O] with respect to [H2 O], in the case [H2 ]o > 2[O2 ]o . We
are assuming that the initial concentration of water is zero, so that, as time goes to infinity,
it reaches the steady state (2[O2 ]o , 0). To produce this graph in Maple, values of [H2 ]o = 5,
[O2 ]o = 2, and k = 0.25 were assumed. Notice that [H2 ]o is stable on the right but unstable on
the left, while 2[O2 ]o is stable on both sides.
Remember that we made the assumption that [H2 O]o = 0. That is, the initial concentration
of water before any chemical reaction occurred was zero. Thus, as time goes to infinity, [H2 O]
approaches 2[O2 ]o . Using limit notation, this is:
lim [H2 O] = 2[O2 ]o (8.19)

t→∞
72
This means, given ample time for the chemical reaction to complete, that the concentration
of water will equal twice the initial concentration of oxygen. Recall that the initial concentration
of oxygen was 2 moles per liter. When the reaction is done, the concentration of water would
be 4 moles per liter (twice the initial concentration of oxygen). Using Equation (8.15), it is
easily verified that when the reaction is done, the concentration of hydrogen would be 1 mole
per liter. Also, using Equation (8.16), it can be found that when the reaction is completed, the
concentration of oxygen would be 0 moles per liter. Thus, the reaction uses up all 2 moles of
oxygen, but only 4 moles of hydrogen, to produce 4 moles of water per liter.
A final observation that is important is that the initial rate of change of the concentration
of water with respect to time is exactly the quantity 2k[H2 ]2o [O2 ]o . As time increases, the rate of
change of the concentration of water with respect to time decreases steadily until all the oxygen
is consumed, at which point it is zero. Using the quantities of [H2 ]o = 5, [O2 ]o = 4, and k = 0.25
(units omitted), the quantity above is 25. Now, we may use Equations (8.14), to find the initial
rate of change of the other two concentrations with respect to time. The initial rate of change
of the concentration of hydrogen with respect to time is exactly −25, while the initial rate of
change of the concentration of oxygen with respect to time is exactly −12.5. Furthermore, since
oxygen and hydrogen are reactants, the concentration of those chemicals with respect to time
will steadily increase until the all the oxygen is consumed, at which point the rates become zero.
We could, of course, create graphs for d[O

dt
2]
and d[Hdt
2]
if we derived equations for them, but
that would be very redundant. All of the information above can be deduced from any one of
the three graphs and any one of the three equations. Producing two or more graphs would not
give us much more information about the model. However, it may be a good exercise for the
reader to produce the graphs and equations for d[O dt
2]
and d[H dt
2]
and interpret them, and ensure
that they do indeed yield the same conclusions above.
8.4.2 Case 2: [H2 ]o < 2[O2 ]o

We will now analyze the case in which the initial concentration of hydrogen is less than twice
the concentration of oxygen. More specifically, let C1 = 2 moles per liter and C2 = 5 moles per
2
liter. We may use the same rate constant, k = 0.25 molL2 ·sec . We may use the same procedure as
the previous section. We will use the same axes as before, with u(t) on the x-axis and u̇(t) on
the y-axis. Of course, Equation (8.18) still intersects the x-axis at the points (C1 , 0) and (2C2 , 0)
and intersects the y-axis at (0, 2kC12 C2 ). However, the intervals in which u̇(t) is positive and in
which it is negative are different. If u(t) < 2C2 , then u̇(t) is positive, otherwise it is negative.
Figure (8.2) is the resulting plot of u̇(t), where C1 = 2 moles per liter, C2 = 5 moles per
2
liter, and k = 0.25 molL2 ·sec .
Recall that we are assuming that the initial concentration of water is zero, so [H2 O]o = 0
moles per liter. Thus, as time goes to infinity, [H2 O] approaches [H2 ]o . In limit notation:
lim [H2 O] = [H2 ]o (8.20)

t→∞
We can also deduce from the graph that the initial rate of change in the concentration
of water with respect to time is 2k[H2 ]2o [O2 ]o , which happens to be 10 moles per liter·time
73
Figure 8.2: This is a graph of d[Hdt2 O] in the case where [H2 ]o < 2[O2 ]o . More specifically, this
is a graph in which the initial concentration of hydrogen, [H2 ]o , is 2 moles per liter, the initial
2
concentration of oxygen, [O2 ]o , is 5 moles per liter, and the rate constant, k, is 0.25 molL2 ·sec .
Notice that [H2 ]o is unstable on the right but stable on the left, while 2[O2 ] is stable on both
sides.
for our specific values for [H2 ]o , [O2 ]o , and k. As time increases, the rate of change in the
concentration of water steadily decreases until the concentration of water reaches the initial
concentration of hydrogen. We can use Equations (8.14) to find out the initial rate of change
in the concentration of hydrogen and oxygen are with respect to time. , d[H dt
2]
= −k[H2 ]2o [O2 ]o
d[H2 ]
and dt = −2k[H2 ]2o [O2 ]o , which is what we got in the other case. Since oxygen and hydrogen
are the reactants, we can expect that the rate of change in concentrations will approach zero
as the chemical reaction progresses. Also, since the chemical reaction produces [H2 ]o = 2 moles
per liter of water, we can use Equations (8.15) and (8.16) to deduce that the chemical equation
requires 1 mole per liter of O2 and 2 moles per liter of H2 . Thus, all of the hydrogen is consumed
when the chemical reaction is complete, but 4 moles per liter of oxygen remain.
8.5 Autocatalytic Reactions

Sometimes, a molecule can be its own catalysis. A catalysis is a reactant that speeds up or
encourages the production of the products(s). In an autocatalytic reaction, a molecule is found
as a reactant and as a product. Autocatalytic reactions are found all over biology. For example,
proteins, which are translation products, enhance the transcription rate of their own genes. The
product are messenger RNA. In turn, messenger RNA increase the translation reaction, creating
more proteins. Thus, proteins act as a reactant and as a product.
74
8.5.1 Example 1
The simplest autocatalytic reaction is:
k1
A + X 2X (8.21)
k−1
In this chemical reaction, molecules of species A reacts with molecules of species X in such
a way that A converts to X to produce 2X. We will assume that the reverse reaction, in which
2X converts to A and X, is also possible. To keep the model simple, we can assume that the
concentration of molecules A, [A], is constant. So we will only derive the equation for d[X]
dt
.
Note that although the forward and backward reactions are possible here, we consider this
as only one chemical equation. Thus, P = 1. Also:
N =2 X1 = A X2 = X V11f = 1 V21f = 1 V11r = 0 V21r = 2
With these values, we can readily calculate q1 , as follows, where [X] and [A] are the concen-
trations of X and Y , respectively, in moles per liter.
q1 = k1 [A]1 [X]1 − k−1 [A]0 [X]2

q1 = k1 [A][X] − k−1 [X]2
Now, we can easily derive the equation for the rate of change in the concentration of X:
d[X]
= (V21r − V21f )q1 = q1 (8.22)
dt
= k1 [A][X] − k−1 [X]2 (8.23)
1
Using dimensional analysis, we can conclude that the units for k1 are time and the units for
L
k−1 are mole·time . Notice that the units for these rate constants are different from the examples
in the previous section.
Next, we may proceed with the graph of d[X] dt
, using the same procedure that we used in the
previous two sections. Let [X] be on the x-axis and d[X] dt
be on the
y-axis.
It is very easy to
verify that the graph intersects the x-axis at the points (0, 0) and kk1−1 [A]
, 0 (you may want to
∗
check this yourself). For simplicity, let us denote the latter point as (x , 0). Note that [X] > 0
within the interval (0, x∗ ) and negative elsewhere. Also note that Equation (8.23) is quadratic,
so it should be easy to graph.
With this information, we can sketch a graph of d[X]dt
. Figure (8.3) shows a graph produced
1 L
by Maple, using (arbitrarily chosen) values of k1 = 0.5 time , k−1 = 0.1 mole·time , and [A] = 5 moles
per liter. Recall that we assumed that [A] is constant, so we can choose its value.
Observe that since the concentration of a molecule cannot be negative, the portion of the
graph displaying negative values for [X] have no chemical meaning. Only the first and fourth
75
1 L
Figure 8.3: This is the graph of Equation (8.23), where k1 = 0.5 time , k−1 = 0.1 mole·time , and
[A] = 5 moles perliter. Notice
that as time increases, the concentration of X eventually reaches
k1 [A]
the steady state k−1 , 0 , no matter what the initial concentration of X is.
quadrants are meaningful. We can see from the graph that, no matter what the initial con-
centration of [X] is, as the reaction progresses, the concentration of X will always approach
(X ∗ , 0). Thus, [X] restricts its own production, although molecule A is always present. Thus
is similar to the chemostat model in which the density of bacteria never exceeds kmax or the
population of a species growing logistically, which never exceeds the carrying capacity.
Explicit Solution for [X](t)

Since [A] is constant, we are fortunate enough to be able to solve explicitly for [X](t) using
Equation (8.23) by separation of variables. The solution is:
k1 [A]ek1 [A](t+C)
[X](t) = (8.24)
1 + k−1 ek1 [A](t+C)
Now, we are able to identify what the concentration of X is at any time t instead of what
the concentration of X is as time goes to infinity. Figure (8.4) shows a graph of Equation (8.24),
where C = −1, k1 = 0.5, k−1 = 0.1, and [A] = 5 (units omitted). Notice that the maximum
concentration of X is X ∗ = kk1−1[A]
, as expected. Of course, negative values of [X] have no
chemical meaning in the graph, and should be ignored. The value of C depends on the initial
concentration of X.
76
Figure 8.4: This is a graph of Equation (8.24), where C = −1, k1 = 0.5, k−1 = 0.1, and [A] = 5.
With this graph, you can determine the concentration of X at any time t. Notice that the
maximum concentration of X is kk1−1
[A]
, which is consistent from our analysis of Figure (8.3).
Bifurcation Diagram for [X](t)

A third type of graph that can be produced is called a bifurcation diagram. Just like in the
other two graphs, a bifurcation diagram can be used to determine the stability of a differential
equation. A parameter is plotted along the x-axis and the dependent variable is plotted along
the y-axis. Then, curves are plotted on the graph wherever the derivative equals zero. On these
curves, the system can either be stable or unstable. These curves are analogous to the nullclines
in phase plane diagrams.
To create a bifurcation diagram of Equation (8.23), we need to find out when the derivative
equals zero, as follows:
d[X]
= k1 [A][X] − k−1 [X]2 (8.25)
dt
k1 [A]
0 = k1 − [X] [X] (8.26)
k−1
k1 [A]
Let µ = . Then: (8.27)
k−1
= k1 (µ − [X])[X] (8.28)
It can be easily verified that [X] = 0 and [X] = µ are always solutions where the derivative
equals zero. The next step is to determine if the differential equation is stable or unstable on
each line. Just like nullclines, the stability may only changes when the two lines intersect. A
77
differential equation is stable on a line if solutions approach it on both sides, and unstable if
solutions avoid it on both sides. For example, look at Figure (8.5). At point A, µ > [X] > 0, so
the derivative is positive. Therefore, as [X] increases, the solution approaches the line [X] = µ.
Similarly, at point B, [X] > µ > 0, so the derivative is negative. Therefore, as [X] increases,
the solution also approaches the line [X] = µ. Since solutions approach [X] = µ on both sides
when µ > 0, it the differential equation is stable on [X] = µ. The stability for [X] = µ changes,
however, when it crosses the origin, because it intersects with the line [X] = 0.
Figure 8.5: This is a bifurcation diagram of Equation (8.23). The blue lines are the where the
differential equation is stable, the red lines are where the differential equation is unstable, and
the green lines are direction fields. Notice that when the concentration of X is positive, µ is
always stable. Thus, as the chemical reaction progresses, the concentration of X approaches µ.
What this bifurcation diagram shows is consistent with what was mentioned before. If
[X] > 0 (which it must be), then the line [X] = µ is stable. Thus as time increases, the
concentration of X will approach µ = kk1−1
[A]
.
8.5.2 Example 2
The first example was a very simple chemical equation with autocatalysis. This example will
deal with two chemical equations with autocatalysis, shown below:
78
k1
A + X 2X (8.29)
k−1
2 k
B + X →C (8.30)
In this example, the first chemical equation is reversible, but the second one is not. To
simplify this example, the assumption will be made that [A] and [B] are constant. We have:
P =2 N =4 X1 = A X2 = X X3 = B X4 = C
Assignments of X1 through X4 are arbitrary. You could, for example, choose X1 = X

instead, but the reaction rate law will end out being the same. Using the chosen values of X1
through X4 above, we can find the values for V f and V r .
V11f = 1 V21f = 1 V31f = 0 V41f = 0

V12f = 0 V22f = 1 V32f = 1 V42f = 0
V11r = 0 V21r = 2 V31r = 0 V41r = 0
V12r = 0 V22r = 0 V32r = 0 V42r = 1
Written in matrix notation, V f and V r are:
   
1 0 0 0
1 1 2 0
Vf =
0
 Vr = 
1 0 0
0 0 0 1
Now, we may calculate q1 and q2 :
q1 = k1 [A]1 [X]1 [B]0 [C]0 − k−1 [A]0 [X]2 [B]0 [C]0

= k1 [A][X] − k−1 [X]2
q2 = k2 [A]0 [X]1 [B]1 [C]0 − 0[A]0 [X]0 [B]0 [C]1
= k2 [X][B]
Finally, we can derive an equation for the rate of change in the concentration of [X]. We
do not need equations for d[A]
dt
or d[B]
dt
because they are zero (we assumed that [A] and [B] are
constant), and we do not need an equation for d[C]
dt
because, as we will soon see, d[X]
dt
does not
d[X]
depend on [C]. The equation for dt is:
79
d[X]
= R2 = (V21r − V21f )q1 + (V22r − V22f )q2 (8.31)
dt
= (2 − 1)q1 + (0 − 1)q2 (8.32)
= q1 − q2 (8.33)
= k1 [A][X] − k−1 [X]2 − k2 [X][B] (8.34)
= [X](k1 [A] − k2 [B] − k−1 [X]) (8.35)
We can simplify this equation slightly by introducing a new parameter, λ.

Let λ := k1 [A] − k2 [B]. Then:
d[X]
= [X](λ − k−1 [X]) (8.36)
dt
Note that λ is a constant, and that λ can either be positive or negative. Also notice that
Equation (8.36) is just the logistic equation u̇ = u(λ − βu)!
Since λ can either be positive or negative, we have two cases. First, let us consider the
positive case. We can observe that Equation (8.36) is quadratic and is concave downward.
Thus, it is fairly simple to sketch a graph similar to Figure (8.6).
Figure 8.6: This is the graph of Equation (8.36) in the case in which λ > 0. For this plot,
arbitrary values of k1 = 0.5, k−1 = 0.1,k2 = 0.2,
[A] = 5, and [B] = 2 were chosen. Notice that
λ
(0, 0) is an unstable steady state, but k−1 , 0 is a stable steady state.
Since negative values for [X] do not make sense, we will only concern ourselves with the first
and fourth quadrants. It is apparent from Figure (8.6) that when λ > 0, the concentration of
80
λ
X always approaches k−1
as time increases. Also, note that [X] always avoids the origin. Thus

λ
(0, 0) is an unstable steady state, while k−1 , 0 is a stable steady state.
Now, lets observe what happens when λ < 0. Both steady
states exist, and Equation (8.36)
λ
is still quadratic and concave downward, but now k−1 , 0 is on the negative x-axis instead
of the positive x-axis. It is very easy to sketch a graph of this case. The result is similar to
Figure (8.7).
Figure 8.7: This is the graph of Equation (8.36) in the case in which λ < 0. For this plot,
arbitrary values of k1 = 0.5, k−1 = 0.1, k2
= 0.2, [A] = 2, and [B] = 10 were chosen. Notice that
λ
(0, 0) is a stable steady state, but k−1 , 0 is an unstable steady state.
From Figure (8.7), it is clear that as time increases,

the concentration
of X approaches the
λ
origin. Thus, the origin is a stable steady state and k−1 , 0 is an unstable steady state. We

λ
should be happy that k−1 , 0 is not stable because if it were, it would imply that [X] will
approach that steady state, which is in the negative x-axis. We do not want [X] to be negative!
It is interesting to note that when λ < 0, the chemical equations (8.29) and (8.30) do not
support the production of X, and so X will actually be consumed entirely. Thus, λ creates
a threshold effect. If λ > 0, then the chemical equations (8.29) and (8.30) will support the
λ
presence of X, and eventually, the concentration of X will stabilize to k−1 . If λ < 0, then what
was described above will occur.
Interestingly, Equation (8.36) can be solved explicitly for [X]. However, the procedure is
much more tedious than in Example 1. Using an computer algebraic system (such as Maple),
the solution for [X] at any time t can be computed. The solution is:
81
k1 [A] − k2 [B]
[X](t) = −(k [A]−k2 [B])t (k [A]
(8.37)
k1 + Ce 1
1 − k2 [B])
The graph of Equation (8.37) is similar to the graph in Figure (8.4). There is a horizontal
λ
asymptote at [X] = 0 when λ < 0 and at [X] = k−1 when λ > 0.
82
Chapter 9
The Enzyme-Substrate Model
9.1 Introduction
An enzyme is a protein that acts like a catalyst for a biochemical reaction. A substrate is a
molecule in which an enzyme acts upon. In an enzyme substrate reaction, the substrate attaches
to the enzyme. This causes a chemical reaction to occur so that the substrate is transformed
into one or more products. The products are then detached from the enzyme. The enzyme
may continue to collect more substrates. An example of an enzyme substrate reaction is the
chemical decomposition of hydrogen peroxide. Hydrogen peroxide is a substrate that attaches
to the enzyme catalase. The bond transforms hydrogen peroxide into two products, water and
oxygen which separate from the enzyme after they are produced. The enzyme is not changed
in any way during this process.
9.2 The Enzyme-Substrate Model

The following chemical equation models the enzyme substrate reaction in the case where one
substrate attaches to an enzyme and transforms into one product:
k1 k
2
E + S C →E +P (9.1)
k−1
In the above chemical equation, E is the enzyme, S is the substrate, P is the product that
the substrate transforms into, and C is the stage when the substrate is attached to the enzyme.
Notice that the first reaction is reversible, so it is possible that the substrate detaches from
the enzyme without the substrate turning into a product. The second reaction is, on the other
hand, not reversible, so once a substrate turns into a product, the enzyme cannot turn it back
into a substrate.
Using the equations for mass action kinetics in the beginning of this chapter, we can derive
the following system of differential equations. For convenience, we will drop the bracket notation
we have been using to represent the concentration of the molecules and proteins. Thus S is to
be interpreted as [S], E as [E], etc. The steps will be omitted here, but the reader is encouraged
to perform the necessary calculations to get the following model equations:
83
dS
= −k1 SE + k−1 C (9.2)
dt
dE
= −k1 SE + k−1 C + k2 C (9.3)
dt
dC
= k1 SE − k−1 C − k2 C (9.4)
dt
dP
= k2 C (9.5)
dt
In addition to these differential equations, we have the following initial conditions:
S(0) = So E(0) = Eo C(0) = 0 P (0) = 0 C + E = constant = Eo (9.6)
The last initial condition can be explained as follows. In total, there are the enzymes that
are not attached to a substrate and enzymes that are attached to a substrate. Hence, the total
number of enzymes is C +E. By adding Equations (9.3) and (9.4), it can be shown that C +E is
constant. Using this assumption, we can reduce the necessary equations. We will only consider
S and C from this point onward. Substituting E = Eo − C for the corresponding differential
equations, we get:
dS
= −k1 S(Eo − C) + k−1 C = −k1 Eo S + (k1 S + k−1 )C (9.7)
dt
dC
= k1 S(Eo − C) − (k−1 + k2 )C = k1 Eo S − (k−1 + k2 + k1 S)C (9.8)
dt
To simplify the above presentation and to eliminate some unnecessary parameters, we may
wish to nondimensionalize the system. We begin by separating all of the variables into a scalar
multiple component and a unit carrying component, as demonstrated in the table below:
Measured Quantity = Scalar Multiple × Unit Carrying Dimension

S = S∗ × Ŝ
C = C∗ × Ĉ
t = t∗ × τ
Our goal is to rewrite the system of equations so that they no longer have any unit carrying
variables or parameters. In other words, no hats or taus (τ ) should appear in our final form.
But first, we must substitute the variables on the left of the table with the variables on the right
of the table, as shown below:
d(S ∗ Ŝ)
= −k1 Eo S ∗ Ŝ + (k1 S ∗ Ŝ + k−1 )(C ∗ Ĉ) (9.9)
d(t∗ τ )
d(C ∗ Ĉ)
∗
= k1 Eo S ∗ Ŝ − (k−1 + k2 + k1 S ∗ Ŝ)(C ∗ Ĉ) (9.10)
d(t τ )
84
To make the left hand side of the equations above unitless, we can divide by Ŝ in the first
equation and by Ĉ in the second equation, and multiply by τ in both equations:
dS ∗ ∗ ∗ C ∗ Ĉτ
= −k 1 E o τ S + (k 1 S Ŝ + k−1 ) (9.11)
dt∗ Ŝ
∗ ∗
dC k1 Eo τ S Ŝ
∗
= − (k−1 + k2 + k1 S ∗ Ŝ)C ∗ τ (9.12)
dt Ĉ
Now, we are ready to reduce the number of parameters present. Let:
1
τ= Ŝ = So Ĉ = Eo (9.13)
k1 Eo
In the above substitutions, notice that the left hand side are all variables, while the right
hand side are all parameters. Care must be taken to ensure that all substitutions contain
equivalent units. It is left as an exercise for the reader to verify that the units are indeed
equivalent. Making the above substitutions yields:
dS ∗ ∗ ∗ C∗
= −S + (k1 S So + k−1 ) (9.14)
dt∗ k1 So
∗ ∗
dC S So ∗ C∗
= − (k −1 + k 2 + k 1 S So ) (9.15)
dt∗ Eo k1 Eo
We have removed all hats and taus from out system, and it is indeed nondimensionalized.
However, we can simplify the system of differential even more by equating complex expressions
with new parameters, as such:
Eo
Let = So
. Then:
dS ∗

∗ ∗ k−1
= −S + S + C∗ (9.16)
dt∗ k1 So
dC ∗

∗ k−1 + k2
∗ =S − ∗
+ S C∗ (9.17)
dt k1 So
Finally, let λ = k2kS2 o and K = k−1 +k2
k1 So
. Then, the final form of the system of differential
equations is (removing the unnecessary stars):
dS
= −S + (S + K − λ)C (9.18)
dt
dC
= S − (K + S)C (9.19)
dt
Along with the initial conditions:

S(0) = 1 C(0) = 0 (9.20)
85
9.3 Perturbation Method
Finding an explicit solution for S and C in Equation (9.18) is not possible. When it is hard
to derive explicit solutions for a system of differential equations, perturbation methods can be
used. A perturbation method is a numerical technique that is used to solve an approximate
solution of a system of differential equations in lieu of finding the exact solution.
First, let u = S and v = C. Then the enzyme substrate model becomes:
u̇ = −u + (u + K − λ)v (9.21)
v̇ = u − (K + u)v (9.22)
u(0) = 1, v(0) = 0 (9.23)
Recall that = ESoo . Typically, the amount of enzyme present in a reaction is much smaller
than the amount of substrate. Therefore, 0 < 1. As a first approximation, will substitute
= 0 into the system of differential equations. We will see momentarily that this is a rather
crude approximation and we will improve it later.
When = 0, Equation (9.22) becomes:
u
v= (9.24)
K +u
Substituting this into Equation (9.21) yields:
(u + K − λ)u
u̇ = −u + (9.25)
K +u
λ
= −u + 1 + u (9.26)
K +u
du −λu
= (9.27)
dt K +u
Now, we can explicitly solve for u by separation of variables:
K +u
dt = −λdt (9.28)
u
Integrating, K ln u + u = −λt + C (9.29)
At t = 0, K ln 1 + 1 = 0 + C (9.30)
Thus, C=1 (9.31)
The solution for u is given implicitly as:
K ln u + u = 1 − λt (9.32)
86
u(0) 1
However, = 0 is a poor approximation to solve v because v(0) = K+u(0) = 1+K 6= 0, but
the initial condition in (9.23) says that v(0) = 0. The true graph of C(t) starts at C(0) = 0, but
then increases rapidly toward a steady state. Then C(t) gradually declines as time increases.
When we set = 0, this rapid increase is lost, and so we get the incorrect initial condition.
Figure (9.1) illustrates the graph of C(t).
Figure 9.1: This is the graph of C(t). Notice that using the assumption that = 0 is a poor
choice for the inner region because the initial condition C(0) = 0 will be violated, and the graph
1
will start at v(0) = 1+K 6= 0 instead.
Notice that Figure (9.1) consist of two parts, an inner region and an outer region. The inner
region will disappear if = 0, but the outer region is not greatly affected. Also, notice that
in the inner region, dC dt
is large, so dC
dt
has order 1 and it is not approximately 0. Thus, the
equations that we derived are acceptable for the outer region, but not for the inner region.
We have to write two new equations for the inner region of the enzyme-substrate model. We
can accomplish this using a change of variable.
Let τ = t , and let:
U (τ ) = u(t) V (τ ) = v(t) (9.33)
Then:
dU du dt
= = u̇ (9.34)
dτ dt dτ
dV dv dt
= = v̇ (9.35)
dτ dt dτ
87
The enzyme-substrate model becomes:
U 0 = (−U + (U + K − λ)V ) (9.36)

V 0 = U − (K + U )V (9.37)
Now that the change in variable has been made, we can set = 0. Doing so yields:
U 0 (τ ) = 0 (9.38)
V 0 (τ ) = C2 − (K + C2 )V (9.39)
The equation for V (τ ) is a linear first order non-homogeneous differential equation that can
be solved explicitly for V (τ ). Recall that a differential equation in the form a dfdx
(x)
+bf (x)+c = 0,
− ab x
where a, b, and c are constants has a general solution Ae and a particular solution − cb . So,
C
the particular solution for V (τ ) is K+C and the general solution is C1 e−(K+C)τ . Thus, we have:
U (τ ) = C2 (9.40)
C2
V (τ ) = + C1 e−(K+C2 )τ (9.41)
K + C2
U (0) = 1 −→ C2 = 1 (9.42)
1 1
V (0) = 0 −→ + C1 = 0 −→ C1 = − (9.43)
K +1 K +1
1 1
∴ V (τ ) = − e−(K+1)τ and U (τ ) = 1 (9.44)
K +1 K +1
Equations (9.44) are the approximate solutions of U and V in the inner region. Recall that
the outer solution is:
K ln u + u = −λt + C (9.45)
u
v= (9.46)
K +u
We still have to derive the value of C so that the inner and outer regions match. To find
C, let t → 0 in the equations for u and v and let τ → ∞ in the equations for U and V . Since
= 0 in the inner solution, we want τ → ∞ since τ = t/. Then:
lim (U (τ ), V (τ )) = lim (u(t), v(t)) (9.47)

τ →∞
t→∞
1
1, = (u(0), v(0)) (9.48)
K +1
This implies that C = 1 in Equation (9.45). Notice that Equation (9.46) is satisfied.
Finally, an approximate solution of Equations (9.21)-(9.23) consists of an inner solution and
an outer solution. The inner solution is given by:
88
U (τ ) = 1 (9.49)
1 −(K+1)τ

V (τ ) = 1−e (9.50)
K +1
and the outer solution is given by:
K ln u(t) + u(t) = λt + 1 (9.51)

u
v(t) = (9.52)
K +u
We can continue this procedure by finding higher order approximations of the solution. We
have computed the first terms of the following sums:
∞
X
u(t) = n un (t) (9.53)
n=0
X∞
v(t) = n vn (t) (9.54)
n=0
∞
X
U (τ ) = j Uj (τ ) (9.55)
j=0
X∞
V (τ ) = j Vj (τ ) (9.56)
j=0
If we wish to have more accuracy, then we can compute more terms in the above sums.
9.4 Problems
1. (Enzyme-Substrate-Inhibitor Model) Consider the following chemical reaction network.
The part on the left is the enzyme substrate model discussed in this chapter. The part
on the right represents an inhibitor I competing for the enzyme E to produce a second
complex D. The symbols on top of the left and right headed arrows are the reaction rate
constants.
k1 k k3
2
E + S C →E +P E+I D
k−1 k−3
Using the law of mass action, write down a system of ODEs that describes the dynamics
of the chemicals E, S, C, P, I and D. Reduce your system to as small a set of equations
as possible. You may assume that E(0) = Eo , S(0) = So , C(0) = 0, P (0) = 0, I(0) = Io ,
and D(0) = 0. No analysis or non-dimensionalization is necessary.
89
2. (Cancer Model) Let p(t) denote the number of cancer cells and let λ be the immune
response. Without the immune response, the cancer cells grow logistically. With the
immune response, the dynamic of the cancer cells follows the equation:
λp
ṗ = rp(1 − p) −
K +p
where r, K, λ are positive constants.
(a) Find the steady states of this equation and show that there exists a λ∗2 > 0 such that
if the immune response is too strong (λ > λ∗2 ), then the cancer cells will not survive.
(b) If λ < λ∗2 , then there are two steady state of the above equation, which we shall
denote by p∗1 , p∗2 , where p∗1 (λ) ≥ p∗2 (λ). Plot the graphs of these two functions for
the cases K > 1 and 0 < K < 1 separately, with λ on the horizontal axis and p on
the vertical axis. Each graph should look like a parabola facing left. They are called
bifurcation diagrams.
(c) Discuss the behavior of the solutions of the above ODE with initial condition p(0) =
po > 0 for the two cases in part (ii). Consider different possibilities for the values of
po . What can you conclude about the stability of the two branches pi (λ), i = 1, 2?
90
Chapter 10
Limit Cycles and the

Poincaré-Bendixson Theorem
10.1 Introduction
A phase plane for a system of two differential equations frequently contains a cycle. Consider
the following system:
ẋ = f (x, y) (10.1)
ẏ = g(x, y) (10.2)
A cycle is a non-constant periodic solution of Equations (10.1) and (10.2). One example
of a cycle is a periodic solution around a center equilibrium point, such as the point (1,1) in
Figure (3.4). Note that a cycle need not be a circle, but may be any simply connected closed
shape. A simply connected shape is a shape in which sides and edges do not overlap. In a
phase plane, solution curves may not overlap because the Uniqueness Theorem guarantees that
each point may be a solution at only one time t. Occasionally, solution curves do not form a
cycle, but approach a cycle instead. Such cycles are called limit cycles.
Before the Poincaré-Bendixson Theorem is introduced, some terminology must be explained.
Let C + be the set of all points, P (t), in R2 that are solutions to Equations (10.1) and (10.2) for
all t ≥ t0 . Hence C + is an orbit of Equations (10.1) and (10.2) when time goes in the forward
direction. Let C − be the set of all points, R(t), in R2 that are solutions to Equations (10.1)
and (10.2) for all t ≤ t0 . Hence C − is an orbit of Equations (10.1) and (10.2) when time goes
backwards. Then, the point Q in R2 is a limit point of C + if the there exist a sequence of real
numbers, {tn }, where tn → ∞ as n → ∞, such that P (tn ) → Q. Similarly, Q is a limit point
of C − if there exists a sequence of real numbers, {tn }, where tn → − ∞ as n → ∞, such that
R(tn ) → Q. Note that all stable and unstable steady states are limit points. A limit set is the
collection of all limit points Q for either C + or C − . There are two types of limit sets: α and ω.
An ω limit set is the collection of all points Q for a trajectory C + as t → ∞. An α limit set is
the collection of all points Q for a trajectory C − as t → − ∞. Sometimes, α and ω limit sets
contain limit cycles. This occurs when the points Q is a cycle in which C + or C − approaches.
Figure (10.1) illustrates α and ω limit sets that have limit cycles. Note that there is a third
91
Figure 10.1: Limit cycles can be found in α and ω limit sets. An α limit set occurs when solution
curves approach a collection of points Q as t → ∞. An ω limit set occurs when solution curves
approach a collection of points Q as t → −∞. These limit sets contain a limit cycle when Q is a
cycle. In this figure, the red and green trajectories are solution curves, and the blue dotted line
is the cycle the solution curves approach. Note that for the degenerate case, the red solution
curve approaches the cycle in the forward direction (i.e as time goes forward), while the green
solution curve approaches the cycle in the backward direction (i.e as time goes backwards.)
type of limit cycle, the degenerate limit cycle, in which both C + and C − approach a cycle in
both forward and backward time. We will not discuss this special case in this chapter.
10.2 Poincaré-Bendixson Theorem

The following theorem explains the conditions in which an ω limit cycle exist:
Poincaré-Bendixson Theorem. If the autonomous system (10.1)-(10.2) in R2 has a bounded

solution C + that is contained in a simply connected domain D that has a boundary d, and if
D contains no steady states, then there exist a limit cycle for the ω limit set contained in
D := D ∪ d.
An autonomous system of differential equations is a system that does not depend on the
independent variable, which is usually time t. In fact, all of the differential equations that we
have been using are autonomous because the right hand side of the equations do not have t in
them. A bounded solution is a solution that gets ”‘trapped”’ inside D. In other words, once
any solution curve enters D in forwards time, it never comes out in forwards time.
92
10.3 Example 1: Scaled Predator-Prey Model
Consider the following autonomous system of differential equations:
dx 9xy
= x(1 − x) − (10.3)
dt 10x + 1
dy 10y
= 0.3y 1 − (10.4)
dt 10x + 1
This system is a scaled predator-prey model. You may compare this system with Equations (4.4)
and (4.5). The first term in Equation (10.3) is the logistic growth of the prey in the absence
of the predator, where α1 = K = 1. The second term in Equation (10.3) limits the number of
prey that a predator can kill when x is large. Equation (10.4) simplifies to the logistic equation
when x is constant, so that that carrying capacity of the predator population depends on x. To
be precise, the carrying capacity for the predator is (10x + 1)/10.
This system has a limit cycle. To find the domain D in which the limit cycle resides in, we
must first locate all steady states. We must choose D so that it does not contain any steady
states, so locating all the steady states will tell us where the limit cycle is not located. To find
the steady states, we must find where the nullclines intersect.
The x nullclines are:
dx 9x
= 0 = x(1 − x) − (10.5)
dt 10x
+1
9y
0=x 1−x− (10.6)
10x + 1
9x
x=0 or 0=1−x− (10.7)
10x + 1
9xy
(1 − x) = (10.8)
10x + 1
1
y = (1 − x)(10x + 1) (10.9)
9
The y nullclines are:

dy 10y
= 0 = by 1 − (10.10)
dt 10x + 1
10y
y=0 or 0=1− (10.11)
10x + 1
10x + 1 = 10y (10.12)
1
y =x+ (10.13)
10
We can easily find the steady sates algebraically, but if you wish to find them graphically,
you can enter the following in Maple:
93
Figure 10.2: This is the plot Maple produces for the nullclines of the scaled predator-prey model.
Here, the x nullclines are the red curve and the y axis and the y nullclines are the green line
and the x axis. The steady states are the intersections of the x nullclines with the y nullclines.
They are (0,0), (1,0), (0, 0.1), and (0.1, 0.2).
1 1

plot 9
· (1 − x) · (10x + 1), x + 10
, x = 0..1
The first parameter is a list of all the expressions to be graphed. If expressions instead of
equations are put in this list, then Maple assumes that the expressions are in the form ”‘y = ...”’,
so there is no need to put ”‘y =”’. The second parameter is optional. It tells Maple the domain
to plot the graph. In this case, the domain is 0 < x < 1. Figure (10.2) is the graph that Maple
produces from the line of code above. The x nullclines are the red curve and the y axis and the
y nullclines are the green line and the x axis. It is clear from the figure that steady states are
(0,0), (1,0), (0, .1), and (.1, .2).
Let us now construct the domain E, a square that is bounded by x = 0, x = 1.1, y = 0,
and y = 1.1. Figure (10.3) shows this region in blue. We want to show that E is an invariant
region, that is, a region that traps trajectories inside so that once they enter they cannot exit.
We can do this by determining the direction of the direction fields along the boundary of E:
• For x = 0: This is a x nullcline, so the direction fields on this line must be vertical.
Furthermore, since dydt
< 0 when 0 < y < 1.1, the direction fields go directly downward on
x = 0. Since the direction fields go downward, solution curves cannot enter or exit from
this side of the square.
94
Figure 10.3: This figure shows the region E in blue with direction fields on the boundary of
E. Notice that there are three steady states on the boundary of E, but this is not a violation
of the Poincaré-Bendixson Theorem. However, the steady state (0.1, 0.2) is a violation of the
theorem because it is inside of the region. We must create a region that does not contain this
steady state.
• For x = 1.1: dxdt

< 0 for 0 < y < 1.1, so the direction fields go leftward. Since the direction
fields go leftward, solution curves can enter, but not exit, the square on this side.
• For y = 0: This is a y nullcline, so the direction fields on this line must be horizontal.
Furthermore, since dxdt
> 0 when 0 < x < 1, the direction fields go directly rightward on
y = 0. Since the direction fields go rightward, solution curves cannot enter or exit from
this side of the square.
• For y = 1.1: dy
dt
< 0 for 0 < x < 1, so the direction fields go downward. Since the direction
fields go downward, solution curves can enter, but not exit, the square on this side.
It should be clear now that E is indeed an invariant domain because once solution curves
95
enter the region, they cannot exit. Figure (10.3) shows the direction fields on the boundary of
E in red.
Notice that (0, 0), (1, 0), and (0, 0.1) lie on the boundary of E. This does not violate
the Poincaré-Bendixson Theorem because they do not lie inside the region. The steady state
(0.1, 0.2), however, does lie inside E, which is not allowed by the theorem, so we must somehow
modify E so that it no longer contains (0.1, 0.2) and so that it will still be an invariant domain.
Figure 10.4: This is a phase plane diagram near the steady state (0.1, 0.2), surrounded by a
region F . Notice that solution curves cannot enter region F , and may only exit on the vertical
boundary.
To do this, lets investigate the behavior of the solution curves around (0.1, 0.2). Figure (10.4)
shows the phase plane near the steady state. It can be shown (with great difficulty) that (0.1,
0.2) is an unstable spiral. The figure also shows the region F . The boundary of region F is on a
solution curve, except for a vertical line in which the direction fields point rightward. Solutions
cannot enter this region, but may exit by passing this vertical line.
Finally, we may construct an invariant region that does not contain any steady states. Define
D := E −F . The region D is show in Figure (10.5). This is an invariant domain because solution
curves can only enter this region as time goes forward. Solution curves cannot leave this region.
Notice that while solution can only exit region F , solution may only enter region D. Also,
notice that the steady state (0.1, 0.2) that was in regions E and F is not in region D. Since all
the conditions for the Poincaré-Bendixson Theorem are now satisfied, it follows that there exist
a limit cycle in D. Figure (10.5) is the resulting phase plane with a limit cycle in the ω limit
set. Figure (10.6) shows a zoomed portion of the phase plane to show the limit cycle better.
96
Figure 10.5: This is a phase plane diagram of the scaled predator-prey model. Notice that the
region D is an invariant domain that does not contain any steady states, except on its boundary
d. By the Poincaré-Bendixson Theorem, there is an ω limit cycle inside D = D ∪ d.
10.4 Example 2: Converting to Polar Coordinates

Sometimes, it is convenient to convert a system of differential equations into polar form to show
that the system has a limit cycle. Consider, for example, the system:
dx
= x + y − x(x2 + 2y 2 ) (10.14)
dt
dy
= −x + y − y(x2 + 2y 2 ) (10.15)
dt
To convert this system into polar coordinates, recall that:
r2 (t) = x2 (t) + y 2 (t) (10.16)

Differentiating both sides of Equation (10.16), we get:
2r(t)r0 (t) = 2x(t)x0 (t) + 2y(t)y 0 (t) (10.17)

r(t)r0 (t) = x(t)x0 (t) + y(t)y 0 (t) (10.18)
97
Figure 10.6: This is a zoomed portion of Figure (10.5) to better illustrate the limit cycle.The
red dots are the steady states, the pink lines are the x nullclines, and the orange lines are the
y nullclines.
From this point forward, we will eliminate the parameter t in our notation to simply the
proceeding computations. Substituting Equation (10.14) for x0 and Equation (10.15) for y 0
yields:
rr0 = x[x + y − x(x2 + 2y)] (10.19)

+ y[−x + y − y(x2 + 2y)] (10.20)
= x2 + y 2 − x2 (x2 + 2y 2 ) − y 2 (x2 + 2y 2 ) (10.21)
= x2 + y 2 − (x2 + y 2 )(x2 + 2y 2 ) (10.22)
To simplify Equation (10.22), we can use the fact that:
98
y = r sin θ (10.23)
y 2 = r2 sin2 θ (10.24)
y2
sin2 θ = 2 (10.25)
r
y2
sin2 θ = 2 (10.26)
x + y2
Thus, Equation (10.22) simplifies to:
rr0 = x2 + y 2 − x4 − 2x2 y 2 − y 2 x2 − 2y 4 (10.27)

= x2 + y 2 − x4 − 3x2 y 2 − 2y 4 (10.28)
= r2 − (x4 + 3x2 y 2 + 2y 4 ) (10.29)
y2

2 4 2 2 4
= r − (x + 2x y + y ) 1 + 2 (10.30)
x + y2
y2

2 2 2 2
= r − (x + y ) 1 + 2 (10.31)
x + y2
= r2 − r4 (1 + sin2 θ) (10.32)
By dividing both sides of Equation (10.32) by r, we finally have an equation for r0 :
r0 = r − r3 (1 + sin2 θ) (10.33)
To derive an equation for θ, recall that:
y
tan θ = (10.34)
x
Differentiating Equation (10.34) and substituting Equations (10.14) and (10.15) yields:
xy 0 − yx0
sec2 θθ0 = (10.35)
x2
x(−x + y − y(x2 + 2y 2 ))
= (10.36)
x2
y(x + y − x(x2 + 2y 2 ))
− (10.37)
x2
−x + xy − x3 y − 2xy 3
2
= (10.38)
x2
xy + y − x3 y − 2xy 3
2
− (10.39)
x2
2 2
−x − y
= (10.40)
x2
99
To simplify Equation (10.40), we can use the fact that:
x = r cos θ (10.41)
∴ x2 = r2 cos2 θ (10.42)
(10.43)
Equation (10.40) simplifies to:
−r2
sec2 θθ0 = (10.44)
r2 cos2 θ
−1
= (10.45)
cos2 θ
Finally, dividing both sides of Equation (10.45) by sec2 θ, we get the equation for θ0 :
θ0 = −1 (10.46)
Thus, the polar form of Equations (10.14) and (10.15) is:
r0 = r − r3 (1 + sin2 θ) (10.47)
θ0 = −1 (10.48)
Our goal is to find an invariant domain that contains no steady states. Consider the domain
D := {(x, y) : 0.5 < r < 1.1}. This domain is the annulus with inner radius 0.5 and outer radius
1.1. Then:
1 1 1 1
r0 |r=0.5 = − (1 − sin2 θ) ≥ − > 0 (10.49)
2 8 2 4
r0 |r=1.1 = 1.1 − 1.331(1 − sin2 θ) ≤ 1.1 − 1.331 < 0 (10.50)
Thus, direction fields go outward around the inner radius and inward around the outer radius.
Once solution curves go inside D, they cannot escape D. Therefore, D is an invariant region.
In addition, it can be shown by finding the intersection of the x and y nullclines in Cartesian
coordinates that the only steady state is (0,0). You must use Cartesian, not polar coordinates
to find the steady states. This steady state is not in D, so D does not contain steady states.
By the Poincaré-Bendixson Theorem, the region D :={(x, y) : 0.5 ≤ r ≤ 1.1} must have a limit
cycle. The phase plane diagram of Equations (10.14) and (10.15) is illustrated in Figure (10.7),
along with the region D.
100
Figure 10.7: This is a phase plane Equations (10.14) and (10.15). The region D is indicated
in purple. D is an annulus with inner radius 0.5 and outer radius 1.1. The only steady state
is (0,0), which is not in D. In addition, D is an invariant region, and once trajectories enter
D they never come out. Notice that there is a limit cycle in D, in accordance with the the
Poincaré-Bendixson Theorem
101
10.5 Example 3: The Van der Pol Equation
The Van der Pol equation is a nonlinear second order differential equation that is used to model
circuits in vacuum tubes. It can also be used to model the human heartbeat. The Van der Pol
equation is:
ẍ − µ(1 − x2 )ẋ + x = 0, µ>0 (10.51)
µ is a nonlinear damping parameter, and is always greater than zero. It can be shown that
there exist an alpha limit set containing a limit cycle in the Van der Pol equation. Before we do
this, however, it is necessary to rewrite the equation into a system of two differential equations
so that we can produce a phase plane for it. Let ẋ = y. Then, ẏ = ẍ = µ(1 − x2 )y − x.
Therefore, the Van der Pol equations can be equivalently written as:
ẋ = y (10.52)
ẏ = µ(1 − x2 )y − x (10.53)
Next, note that the only steady state in the system of equations above is (0, 0). So, the
domain that we create for the Poincaré Bendixson Theorem cannot contain the origin. The
domain that we construct will be similar to an annulus (donut), except the outer radius is not a
circle. To construct the inner radius, we may convert Equations (10.52) and (10.53) into polar
form. Recall that:
r 2 = x2 + y 2 (10.54)
2rṙ = 2xẋ + 2y ẏ (10.55)
rṙ = xẋ + y ẏ (10.56)
Thus:
rṙ = xy + y(µ(1 − x2 )y − x) (10.57)

= xy + y 2 µ(1 − x2 ) − xy (10.58)
= r2 sin2 θ cos2 θ + r2 µ sin2 θ(1 − r2 cos2 θ) − r2 sin2 θ cos2 θ (10.59)
ṙ = rµ sin2 θ(1 − r2 cos2 θ) (10.60)
Define the inner radius to be L1 : r = 12 . Then:
1 1
ṙ = µ sin2 θ(1 − cos2 θ) (10.61)
2 4
1 1
≥ µ sin θ(1 − cos2 θ) = µ sin2 θ ≥ 0
2
(10.62)
2 2
102
Since ṙ ≥ 0, the trajectories can only exit the inner radius L1 .
The outer radius much more complicated, since it is defined piecewise. The outer radius is
the composition of L21 , L22 , L23 , L24 , L25 , and L26 . The outer radius is constructed as follows.
Note that L24 and L25 are in polar form while the other curves are in Cartesian form.
Let A = 4µ + 1/µ and
p B be equal to the smallest natural number x such that the inequality
2 2
1 + (4µ − x)
p > (1 + x/µ) . Now, let D = A if A > B or let D = B if B > A. Finally, define
C = 4µ + (1 + (4µ − D)2 )2 − 1 + 4µ. Then,
p
L21 : y = −µ(x − 1)2 + C −1 < x < 1 + C/µ (10.63)
p
L22 : y = µ(x + 1)2 − C −1 − C/µ < x < 1 (10.64)
L23 : r = 1 + (4µ − D)2 1 ≤ x < 1 + (4µ − D)2 (10.65)
L24 : r = 1 + (4µ − D)2 −1 − (4µ + D)2 < x ≤ − 1 (10.66)
p
L25 : y = 0 (1 + C/µ) ≤ x ≤ 1 + (4µ − D)2 (10.67)
p
L26 : y = 0 −1 − (4µ − D)2 ≤ x ≤ − (1 + C/µ) (10.68)
For example, consider the case when µ = 1. To √ find the value of D, we must determine the
2 2
smallest value of x such that 1 + (4 − x) > (1 + x) . The smallest natural number for x that
works is x = 8. (Technically, we may choose x to be any real number satisfying the inequality,
but choosing a natural number is less complicated). We want to choose the smallest x satisfying
the inequality because we want our domain to be as small as possible so that we can determine
more precisely the location of the limit cycle. We also need to calculate
√ 4µ + 1/µ,
√ which is 5.
Therefore D = max {8, 5} = 8. We can also calculate C = 4 + 288 = 4 + 12 2. The inner
radius for the domain is L1 : r = 1/2, and the outer radius is:
√ √
q
2
L21 : y = −(x − 1) + 4 + 12 2 −1 < x < 1 + 4 + 12 2 (10.69)
√ √
q
L22 : y = (x + 1)2 − 4 − 12 2 −1 − 4 + 12 2 < x < 1 (10.70)
L23 : r = 17 1 ≤ x < 17 (10.71)
L24 : r = 17 −17 < x ≤ − 1 (10.72)
√
q
L25 : y = 0 1 + 4 + 12 2 ≤ x ≤ 17 (10.73)
√
q
L26 : y = 0 −17 ≤ x ≤ − 1 + 4 + 12 2 (10.74)
Figure (10.8) is a phase plane of the Van der Pol system when µ = 1 along with the domain
identified above in green. Notice that the domain does not contain the steady state (0, 0) since
a circle with radius 1/2 has been removed. For L25 and L26 , y = 0 so ẋ = 0 and ẏ = −x.
So trajectories on L25 are going downward as time increases and trajectories on L26 are going
upward as time increases. By preforming similar computations for the other boundaries, it can
103
Figure 10.8: This is a phase plane for the Van der Pol system when µ = 1. The only steady
state is (0,0) which is removed with a circle with radius 1/2. The domain in green is constructed
in such a way that trajectories may enter, but never exit, the domain as time increases. By the
Poincaré Bendixson Theorem, there must be a limit cycle in the domain.
104
be proven that trajectories may enter, but never exit, the domain as time increases. Thus, by
the Poincaré Bendixson Theorem, there exist an alpha limit set containing a limit cycle in the
domain. Figure (10.9) is a phase plane of the Van der Pol system zoomed onto the limit cycle.
Figure 10.9: This is a zoomed phase plane of Figure (10.8).
Instead of using the Poincaré Bendixson Theorem to prove the existence of a limit cycle
for the Van der Pol equation, a different theorem can be used called Liénard’s Theorem. A
Liénard equation is a differential equation in the form:
ẍ + f (x)ẋ + g(x) = 0 (10.75)
Where f is a continuously differentiable even function and g is a continuously differentiable

odd function. A Liénard equation can be rewritten into the system of differential equations,
called the Liénard system:
ẋ = y (10.76)
ẏ = −f (x)y − g(x) (10.77)
105
Notice that the Van der Pol equation is a Liénard equation. We can use the following
theorem to prove the existence of a limit cycle for the Van der Pol equation and other Liénard
equations.
Liénard’s Theorem. A Liénard system has an alpha limit set containing a limit cycle sur-
rounding the origin if:
• g(x) > 0 for all x > 0

Rz
• limz→∞ 0 f (x) dx = ∞
Rz Rz
• 0 f (x) dx has exactly one positive root, r, where 0 f (x) dx < 0 for 0 < z < r and
monotonic for z > r.
It is left as an exercise for the reader to verify that Liénard’s Theorem applies for the Van
der Pol system. Thus, a limit cycle exists for the Van der Pol system that surrounds the origin.
106
Bibliography
[1] Edelsteirn-Keshet, Leah

Mathematical Models in Biology
Society for Indurstrial and Applied Mathematics, 2005.
[2] ”Lotka-Volterra Model of Competition” Otto, Sarah.

2007. University of British Colombia
Acessed on 14 December 2012.
<http://www.zoology.ubc.ca/∼bio301/Bio301/Lectures/Lecture19/Overheads.html>
[3] ”Nonlinear Two Dimensional Dymanics: Mathematical Appendix” Wiens, Elmer G.

4 October 2012. Egwald Web Services
<http://www.egwald.ca/nonlineardynamics/mathappendix.php<
[4] ”ODE Software for MATLAB” Polking, John C.

23 April 2002. Rice University
<http://math.rice.edu/∼dfield/>.
[5] Wang, Dongping; Zhou, Shangbo; Yu, Juebang

The Existence of Closed Trajectory in the van der Pol Oscillator
Sichuan University, 2002.
107

MQP Ryl-1213

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MQP Ryl-1213

Uploaded by

Copyright:

Available Formats

Project Number: MA-RYL-1213

Math Modeling for Undergraduates

A Major Qualifying Project

submitted to the Faculty

WORCESTER POLYTECHNIC INSTITUTE

in partial fulfillment of the requirements for the

Degree of Bachelor of Science

December 14, 2012

Professor Roger Y. Lui

1 The Simple Pendulum Model 5

2 The Escape Velocity Problem 14

3 The Lotka-Volterra Equations 17

4 Extension on the Lotka-Volterra Equations 27

6 Models for Infectious Diseases 48

7 The Chemostat Model 57

9 The Enzyme-Substrate Model 83

10 Limit Cycles and the Poincaré-Bendixson Theorem 91

The Simple Pendulum Model

1.2 The Model

cos (a − b) = [cos a cos b − sin a sin b]

We are able to rewrite the equation in such a way that

This equation can be rewritten as

Lθ̈ + g sin(θ) = 0 (1.10)

More concisely it can be written as

1.3 The Equations

We begin by stating the following characteristic equation:

This can be written as:

eıx = cos(x) + ı sin(θ) (1.18)

We can use this to express Equation (1.17) in the following manner:

θ(t) = C1 (cos(ω0 t) + ı sin(−ω0 t)) + C2 (cos(−ω0 t) + ı sin(−ω0 t)) (1.19)

The arbirtrary constants C1 and C2 can be rewritten as

θ(t) = D1 cos(ω0 t) + D2 sin(ω0 t) (1.21)

Taking the derivative of our θ(t) we have the equation that

θ̇ = −D1 ω0 sin(ω0 t) + D2 ω0 cos(ω0 t) (1.23)

Now we can use our other intial value to show that:

Finally we can arrive at a precise equation for θ(t) that

λ2 + 2bλ + ω02 = 0 (1.26)

Which has characteristic roots:

θ(t) = C1 eλ1 t + C2 eλ2 t (1.30)

Here C1 and C2 are constants similar to those in Equation (1.29)

1.4 Phase Plane Analysis with Direction Fields

1.6 Related Questions

The Escape Velocity Problem

2.2 The Equations

Evaluating the integral we have

2.4 Conclusions and Questions

The Lotka-Volterra Equations

3.2 The Lotka-Volterra Equations

3.3 Steady States and their Stability

(x1 , y1 ) = (0, 0) and (x2 , y2 ) = ( αγ22 , αγ11 )

For the steady state (0,0):

det(J(0, 0) − λI) = (α1 − λ)(−α2 − λ) = 0

(J(0, 0) − λ1 I)v1 = 0 (J(0, 0) − λ2 I)v2 = 0

Recall that all of the parameters are positive. Therefore, α1 γγ21+α

3.4 Phase Plane Analysis

(a) Find all nullclines of the system.

Extension on the Lotka-Volterra

4.2 A Logistic Modification of the Lotka-Volterra Equa-

The g-nullclines are:

(x1 , y1 ) = (0, 0) (4.15)

The eigenvalues at (x1 , y1 ) are:

det(J(0, 0) − λI) = (α1 − λ1 )(−α2 − λ2 ) = 0 (4.20)

U 0 = (−U + (U + K − λ)V ) (9.36)