Вы находитесь на странице: 1из 130

Mathematics for Biologists

Kirsten ten Tusscher and Alexander Panfilov


Theoretical Biology & Bioinformatics
Utrecht University
i


c Utrecht University, 2011
ii
Contents

1 Preliminaries 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Algebraic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Drawing of functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Vectors, Matrices and Complex Numbers 17


2.1 Scalars, Vectors and Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Matrices and systems of equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3 Differential equations of one variable 33


3.1 Simple differential equations and their solutions . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Qualitative analysis of differential equations . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Systems with parameters. Bifurcations. . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 Extra: Solving linear differential equations . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4 Introduction to systems of two differential equations 45


4.1 Solutions of Linear 2D Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Phase plane, Trajectories and Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Nullclines and Vectorfields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Systems with parameters. Bifurcations. . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.5 Extra: The general solution of linear 2D systems . . . . . . . . . . . . . . . . . . . . . 55
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5 Equilibrium types in systems of two differential equations 59


5.1 Real valued eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Complex valued eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Insufficient information in the phase portrait . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4 Equilibrium types and stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

6 Linear approximation of non-linear 2D systems 71


6.1 Partial derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.2 Linear approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.3 Linearization of a system: Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

7 Efficient analysis of systems of two differential equations 79


iv CONTENTS

7.1 Determinant-trace method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79


7.2 Graphical Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.3 Plan of qualitative analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

8 Limit cycle 91
8.1 Stable and non-stable limit cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.2 Dynamics of a system with a limit cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.3 How do limit cycles occur? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
8.4 Example of a system with a limit cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
8.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

9 Extra Questions 99

10 Dictionary 105

11 Answers for exercises 107


Chapter 1

Preliminaries

1.1 Introduction

The main aim of this first chapter is to refresh your knowledge of high school mathematics and to bridge
the gap between high school mathematics and the mathematics that we will use in the Systems Biology
course. In this course we will frequently have to solve simple algebraic equations and draw of functions.

A difference between the equations and functions used in high school and the ones in this course is the
use of so-called free constants or parameters rather than numbers. As an example, rather than asking
you to solve 0 = (3 − x)(x + 2) (x = 3 or x = −2), we will ask you to solve 0 = (a − x)(x + b) (x = a or
x = −b). Unless explicitly stated differently you may assume that these constants have a positive value.
Another difference is that variables will not necessarily be called x, y or z, but may be the number of
animals in a population N or the concentration of chemicals in a reaction vessel c. As a result you will
have to learn to distinguish which are the variables and which are the free parameters. Remember that
variables are things such as the number of animals in a population of which we would like to understand
the dynamical behaviour. In contrast, a free parameter reflects for example the death rate or birth rate of a
population, its value is constant for a particular population but can be different for different populations.

A problem of using parameters rather than numbers is that you can not use your graphical calculator
to draw functions or solve equations. Of course, you can always fill in some arbitrary numbers for the
parameters and still use your calculator. Note however that the general function or equation may have
more than one solution, whereas by filling in a particular set of made up numbers you will always find
only one of these solutions. As a very simple example, consider having to draw the line y = ax + b.
If we choose a = 1 and b = 2 we get a line with a positive slope and a positive intercept with the x-
axis. However, the general equation y = ax + b also holds for lines with a negative slope and a negative
intercept with the x-axis.

1.2 Algebraic equations

Algebraic expressions

Algebraic expressions may contain numbers, variables, parameters and arithmetic operations. Below, we
2 P RELIMINARIES

review examples of several basic operations which help us to work with algebraic expressions.

One of the most basic algebraic operations is getting rid of parentheses to simplify the expression. For
that we use the following rule:

(a + b)(c + d) = ac + ad + bc + bd

note, that here ac means a ∗ c, etc., as the multiplication sign is often omitted.

Example (open parenthesis): (4x + 2a)(2 − 3x) = 8x − 12x2 + 4a − 6ax

On the other hand, we also often introduce parentheses to decompose an expression into multiple factors
that then can be solved separately:

Example (factor expression): 9x3 + 3x2 − 6a4 x2 = x2 (9x + 3 − 6a4 ) = 3x2 (3x + 1 − 2a4 )

Solving of equations

An equation is a mathematical relationship involving one or more unknown variables. Solving equations
means finding the values of these unknowns such that after substitution of these values back into the
equation the left and right hand sides will be equal to each other. For example: the equation 2x − 16 =
−10 has as a solution x = 3, as 2 ∗ 3 − 16 = 6 − 16 = −10.

Solving equations containing a single value means finding one or more values for the variable for which
the equation holds true. In order to find these solutions we need to move all terms containing the varialbe
to one side of the = sign. This can be achieved by multiplying the left and right hand side of the equation
by the same amount, or by adding or substracting the same amount from the left and right hand side
of the equation. Remember that for quadratic equations of the form ax2 + bx + √ c = 0 we can use the
−b± b2 −4ac
so-called ’abc’ formula, which gives us the two possible solutions as x1,2 = 2a .

Example 1: solve a simple equation with one variable: 0 = 2x − 3 So 2x = 3, so x = 1.5

Example 2: solve a more complex equation with one variable: aN 2 = bN First bring all terms with N to
one side, so aN 2 − bN = 0, than factor out N, so N(aN − b) = 0. Now realize that if A × B = 0 either
A = 0 or B = 0. So, for our equation N = 0 or aN − b = 0. So finally, N = 0 or N = ab .

If an equation contains multiple variables, you can choose which variable you will put on its own to
one side of the = sign. You then solve the equation and express the solution in terms of this particular
variable. Which variable you choose for this is up to you, so it is best to pick the variable for which
solving the equation is easiest. The solution of the equation then describes the value of this variable
as a function of the values of the other variables that were in the equation. Try this with the following
examples:

Example 1: solve the equation for one of the two variables: 0 = aN − bM 2 Note that N is more simple
to solve for than M, so aN = bM 2 and hence N = ba M 2 .

Example 2: solve the equation for one of the two variables: aN = b M N Note that now M is more simple
to solve for than N. First multiply left and right with N. This give us aN 2 = bM and hence M = ba N 2 .

Many complex equations can not be solved analytically. As an example, higher order equations, also
called polynomials, of the form a + bx + cx2 + dx3 + · · · = 0 can only be solved if we can factor them
into multiple terms of lower order that then each can be solved seperately. Sometimes another option is
1.2 A LGEBRAIC EQUATIONS 3

to introduce a new variable y = xn , if all terms of the equation can be somehow rewritten as ayb . This
gives you a simpler equation in y that now can be solved, this solution can then subsequently be filled in
to solve x itself.

Example: solve the following equation: aN 6 − bN 3 = 0. Note that both terms can be written in terms of
N 3 , so we can introduce M = N 3 and rewrite the equation as aM 2 − bM = 0. We already know how to
solve this: M(aM − b) = 0 and hence M = 0 or M = ba . Therefore N 3 = 0 and thus N = 0, or N 3 = ba and
thus N = ( ab )1/3

Working with fractions

When working with fractions there are a number of rules that you should remember.
1. if you multiply a fraction, make sure you always multiply both the numerator (teller) and denominator
(noemer) with the same amount,
2. as long as you multiply or divide numerator and denominator with the same number the value of the
fraction remains the same,
3. you can only add up fractions into a single fraction if the denominators of the fractions are the same,
4. dividing by a fraction is the same as multiplying by the inverse of that fraction.
In formulas we can summarize the above as follows:

a c ac a ca ca a a c ad + bc a c ac
× = , = , = , + = and b
= a× = ,
b d bd b cb cb b b d bd c
b b

a, b, c en d are free parameters. Note that this implies that

a+b a b a a a
= + en 6= + .
c c c b+c b c

2
Example 1: solve the following equation: x+3 = 5 Multiplication of left and right hand sides with x + 3
gives us 2 = 5(x + 3) or 2 = 5x + 15. Bringing 15 to the other side than gives us −13 = 5x and hence
x = −13
5

aN aN
Example 2: solve the following equation: 1+bN − cN = 0 First bring −cN to other side: 1+bN = cN.
2
Next multiply left and right hand sides with 1 + bN: aN = cN + cbN . Now also move aN to the other
side: 0 = (c − a)N + cbN 2 . Finally factor out N: 0 = N((c − a) + cbN). This gives us N = 0 or N = a−c
cb .

Note that not all equations containing a fraction require this “treatment”. As an example, consider the
function f (x) = a − x/b. Here the fraction is x/b of which x is a variable and only b is a parameter. The
function can thus be rewritten as f (x) = a −(1/b)x. If we now define a new parameter as c = 1/b, we can
subsequently rewrite the function as f (x) = a − cx. This function describes a straight line with a slope of
−c = −1/b, intersecting the vertical axis in f (0) = a and intersecting the horizontal axis in 0 = a − cx
which give us x = a/c = ab. Similarly, we can see that the function f (x) = b2ax +4c
is a straight line with a
a
slope of b2 +4c . In summary, if a fraction contains only constants in the denominator and variables in the
numerator we can rewrite such a fraction as a multiplication by a newly defined constant. Note also that
2
a fraction such as f (x) = axx can be rewritten as f (x) = ax, describing a straight line with slope a.

Solving systems of equations

Solving systems of coupled algebraic equations works in principle the same as solving several indepen-
dent equations. However, an important difference is that solutions of one equation are substituted into
other equations to help find their solution.
4 P RELIMINARIES

As an example, consider the equations 0 = 3x + 2y and 3 = y/5. Solving the first equation gives you
3x = −2y and hence x = −(2/3)y. Solving the second equation gives you y = 15. If these two equations
form a system, that is, they are coupled and hence the y in the first and second equation are the same, we
can fill in the y = 15 we obtained from solving the second equation into x = −(2/3)y we obtained from
solving the first equation and obtain x = −10 and hence find the final solution of the first equation. Note
that if you are solving a system of 2 or more algebraic equations it is not necessarily the best approach
to start with the first equation and work your way down. Instead, the smart approach is to start with the
equations that are most easy to solve. Indeed, in the example the second equation is much easier than the
first, and we used it’s solution to find the final solution of the first equation. Furthermore, only fill in a
solution of one equation into a solution of another equation if this actually simplifies matters.

Example 1: solve the following system of equations: 0 = aN − bM and 0 = M(cN − d) From the first
equation we obtain N = ab M. From the second equation we obtain M = 0 or cN − d = 0 and hence N = dc .
If we now combine this with the solution of the first equation, M = 0 gives us N = 0 and N = dc gives us
d b da d da
c = a M and hence M = cb . So we have two solutions N = 0, M = 0 and N = c , M = cb .

an − an2 − bnp = 0

Example 2: solve the following system of equations: Let us start with the second
np − kp = 0
equation, as it looks easier. np − kp = 0 gives us p(n − k) = 0 and hence p = 0 or n = k. Now let us
move back to the first equation, an − an2 − bnp = 0. First, we fill in p = 0. This gives us an − an2 = 0
and hence an(1 − n) = 0 and hence n = 0 or n = 1. This means that from filling in only the first solution
of the first equation we already obtained two overall solutions for this system n = 0,p = 0, and n = 1,
p = 0. Let us also fill in the second solution of the first equation into the second equation, this gives us
ak − ak2 − bkp = 0. Note that as k is a parameter, k = 0 is not a solution!. However, we can divide all
by k, resulting in a − ak − bp = 0 or a − ak = bp and hence p = a−ak b . The third overall solution thus is
n = k, p = b . Summarizing all solutions, we found (0, 0), (1, 0), (k, a−ak
a−ak
b .

1.3 Drawing of functions

Functions of one variable

In science, relationships between quantities are often expressed using functions. The simplest type of
functions are those depending on one variable. A function of one variable is a rule that allows us to find
the value of an “output” variable f from the value of a single “input” variable x. We denote it as f (x).
Below are examples of the most important common functions:

power functions xa , for example


1 √ 1
f (x) = x2 ; f (x) = x 2 = x; f (x) = x−2 =
x2
polynomials, ax3 + .. + cx + d, for example:
f (x) = 3x3 − 2x2 + 1
p(x)
rational functions f (x) = g(x) , for example:
2−x
f (x) =
x2 + 1
trigonometric functions sin, cos, tan:
x
f (x) = 2sin(x); f (x) = cos(2x − 1); f (x) = tan( )
2
1.3 D RAWING OF FUNCTIONS 5

f(x)

x*
x

Figure 1.1:

exponential ax and logarithmic function a log(x)

f (x) = ex+1 ; f (x) =10 log(2x); f (x) = 22x

Derivatives

The derivative of a function f (x) at point x∗ is given by the following limit:

df f (x) − f (x∗ )
f 0 (x∗ ) = = lim∗ (1.1)
dx x→x x − x∗
where the d in d f and dx stands for delta or a change in .... Indeed, a small change in function value d f
divided by the distance along the x-axis over which it occurs dx defines the local slope or derivative of
f (x), f 0 (x), in a point x.
0
The derivative f (x∗ ) denotes the rate of change of a function f (x) at a point x∗ and has many applica-
tions:
• If x(t) is the distance traveled by a car as a function of time t, then dx/dt gives the car’s velocity.
• If n(t) is the size of the population as a function of time, then dn/dt gives the population’s growth rate.

0
Geometrically, the derivative f (x∗ ) gives the slope of the tangent line to the graph of the function at the
given by the following equation:
y = f (x∗ ) + f 0 (x∗ ) ∗ (x − x∗ ) (1.2)
Equation (1.2) is also known as the linear approximation of function f (x) at point x∗ .

Let us check formula (1.2) by approximating the function y = 2x2 + 1 at x∗ = 1. We find: f (x∗ ) =
2 ∗ 12 + 1 = 3, f 0 = 4x, f 0 (1) = 4, hence f (x) ≈ 3 + 4(x − 1). At x = 1.1 this approximate formula gives
f (1.1) ≈ 3 + 4(1.1 − 1) = 3.4. The exact value is f (1.1) = 2 ∗ 1.12 + 1 = 3.42. So the error is just 0.6%.
However, if x = 0, f (0) ≈ 3 + 4(0 − 1) = −1 while the exact value is f (0) = 1. So we see, that the
approximate formula works good if x is close to x∗ only.

Derivatives are important to determine the slope of functions, but also to determine the location of max-
ima and minima of functions, where the derivative of the function equals zero. Using second order
derivatives (the derivative of the derivative) we can even determine the location of bending points of the
function, as here second order derivatives equal zero.
6 P RELIMINARIES

Limits

We call A a limit of the function f (x) when, if x approaches the value a, the value of f (x) gets closer and
closer to A. We write this formally as:
lim f (x) = A (1.3)
x→a

If a function is continuouos, finding the limit is quite trivial: we just need to substitute the value of x = a
into our function f (x):
lim f (x) = f (a) (1.4)
x→a

However, there are several important exceptions.

The first exception to this rule arises when we try to find the limit of the function when x → ∞. Finding
this type of limit is very important as it gives us the asymptotic behaviour of our system when the size
of a population or the level of a concentration becomes very large. Unfortunately, there is no such
number as ’∞’ that we can substitute into our function to find the limit using formula (1.4). For functions
without parameters, we can often guess the limit by substituting very large values in to (1.4), e.g. x =
10000, 20000, etc, but what to do for functions with parameters?

p(x)
This problem is particularly relevant for so called rational functions f (x) = g(x) , where p(x) and g(x) are
both polynomials. In these cases we can always find the value of the limit using the following property
of power functions:
C
lim α = 0 (1.5)
x→∞ x

where C is an arbitrary constant and α > 0.

To understand this, consider that if x approaches ∞ (becomes larger and larger), the power function xα
with α > 0 necessarily also becomes larger and larger, and therefore xCα will get closer and closer to zero,
thus resulting in lim xCα = 0
x→∞

To find a limit using this property we need to do the following: (1) find the highest power of x in our
p(x) p(x)
function g(x) , (2) divide each term in our function g(x) by x to that power, and (3) find the limit of each
separate term using property (1.5). Let us consider three typical examples:

aN 2 −3N
Example (find the limit): lim 2 .
N→∞ 3−2N

aN 2
−3 N2 a− N3
The highest power here is N 2 , division by it gives: N2 N
2 = 3 . The limits of the individual terms
2 −2
3
2 − 2N2 N
N N
a−0
then are: 0−2 = − a2

ax3 −bx2 +c
Example (find the limit): lim 4 , a, b, c 6= 0 .
x→∞ ax −b

ax3 2
ax3 −bx2 +c − bx4 + c4 a b c
x − x2 + x4 0−0+0 0
x4
Similar steps as above give us: ax4 −b
= ax 4
x x
= b = a−0 = a = 0.
4 − b4 a− 4
x
x x

aP−3bP3
Example (find the limit): lim 2 , a, b, c, d 6= 0.
P→∞ cP−dP

aP 3bP3 a
aP−3bP3 − 3 −3b 0−3b −3b
P3 P2
Similar steps as above give us: cP−dP2
= cP
P
2 = c
− Pd
= 0−0 = 0 .
3 − dP3 P2
P P
This final expression does not make sense, since it contains a division by zero. As division by a small
1.3 D RAWING OF FUNCTIONS 7

number results in a large outcome, division by something approaching to zero results in an outcome
approaching infinity. As a consequence, for x → ∞ the function does not approach a finite value.

p(x)
The second non-trivial situation arises when the denominator of our function f (x) = g(x) becomes zero
2
for some value of x, for example f (x) = x−3 for x = 3. In this case the formula (1.4) for limit can also
not be used and a more extensive analysis is necessary. By using our calculator and filling in values
for x that are close to 3 we can observe the following: if x becomes closer and closer to 3 from the
left, e.g. x = 3.1; 3.05; 3.01, 3.005; etc the function value becomes larger and larger, while if x becomes
closer and closer to 3 from the right, e.g. x = 2.9; 2.95; 2.99, 2.995; etc the function value is negative and
2
its absolute value also becomes larger and larger. We can formally write this as lim+ x−3 = +∞, while
x→3
2
lim− x−3 = −∞.
x→3

Limits are very important for drawing functions. Limits for x → ∞ and x → −∞ give us the horizontal
p(x)
asymptotes of our graphs. If we have a function of the form f (x) = g(x) in which the denominator
becomes zero for some value a of x then x → a gives us the vertical asymptotes of our graph.

Important graphs

We usually represent functions using graphs. To do that we plot the value of the variable x along the
x-axis and the value of the function f (x) along the y-axis. Here we have listed typical graph shapes that
are important in this course.

Zero order graphs


y y y y=p
y=2
p
x x x

y=−1

a b c
Figure 1.2:
Equations of the form y = p produce a horizontal line at the level p (fig.1.2).

y x=2 x=−1 y y x=p

x x x

a b c
Figure 1.3:
Equations of the form x = p produce a vertical line crossing the x-axis as x = p (fig.1.3)
8 P RELIMINARIES

Linear graphs

Figure 1.4:
Equations of the form y = ax + p (linear function) produce a straight line with the slope defined by the
parameter a: the larger the absolute value of a, the steeper is the slope (fig.1.4), a positive value of a
results in a graph sloping upward from left to right, a negative value of a results in a graph sloping
downward from left to right. The parameter p in y = ax + p accounts for the vertical shift of the graph
fig.1.4 relative to the x-axis.

Quadratic (second-order) graphs

Figure 1.5:
The equation y = ax2 + bx + c produces a parabola, if a > 0 the parabola is opened upward (dalparabool)
(fig.a), and if a < 0 the parabola is opened downward (bergparabool) (fig.b). The larger the absolute
value of a, the steeper the parabola. The parameter c (fig.1.5c) accounts for the vertical shift of the
graph relative to the x-axis. Similarly, parameter b causes a horizontal shift of the parabola relative to
b
the y-axis. We can compute that the horizontal shift of the parabola is given by − 2a (fig.1.5d). We
can do this by determining the location of the vertex of the parabola, which is a point of extremum
(maximum or minimum) of the function. At this point the derivative of the function equals zero, so
(ax2 + bx + c)0 = 2ax + b = 0, Thus the x coordinate of the vertex is given by xv = − 2a b
, or in other words
b
the (vertex of) parabola is shifted by xv = − 2a from its central location in (fig.1.5c). Note also, that a
parabola may have up to two points of intersection of the graph with the x-axis. These can be found from
the ’abc’ formula for roots of the equation ax2 + bx + c = 0, and if these roots (p, q) are known, the graph
can easily be depicted using them (fig.1.5e). Note, that in this case the vertex of the parabola is always
located at the middle between these two roots. Also note that if the discriminant D = b2 − 4ac of the
abc-formula is > 0 we indeed have 2 intersection points with the x-axis with the parabola vertex in the
middle, if D = 0 we have only one point at which the parabola touches the x-axis which corresponds to
the vertex of the parabola, whereas if D < 0 the parabola does not intersect or touch the x-axis.
1.3 D RAWING OF FUNCTIONS 9

Cubic (third-order) graphs

y=−x3 y y=x3 y y
y=ax3 +bx 2 +cx +d y=a(x−p)(x−q)(x−r)

x x p q r x

a b c
Figure 1.6:
For a general cubic function y = ax3 + bx2 + cx + d we have many more possibilities, which we will
not all discuss here. The two simplest forms are given by the functions y = x3 and y = −x3 depicted in
Fig.1.6a. Important here is the asymptotic behavior of the function at x → ±∞. For y = x3 we see that
y goes to +∞ when x increases and to −∞ when x decreases; for y = −x3 we have exactly the opposite
situation. A general graph of the form y = ax3 + bx2 + cx + d may have up to three zeros that can be
found from the solution of the equation ax3 + bx2 + cx + d = 0, and up to two extrema: one maximum
and one minimum (fig.1.6b). The extrema are points where the derivative of the function is zero, which
in this case results in the following quadratic equation: (ax3 + bx2 + cx + d)0 = 3ax2 + 2bx + c = 0. If the
zeros of the function (p, q, r) are known and the asymptotic behavior is known, the graph can easily be
drawn as shown in Fig.1.6c.

Graphs of power functions

y y y y= x
3

x=y 2 y= x

x x x

a b c
Figure 1.7:
Three examples of graphs of the power function xa involving 0 < a < 1 are shown in Fig.1.7. As 0 < a < 1
the function increases slower than the function y = x and is concave downward (in the first quadrant).

To draw graph y = x let us first draw function x = y2 . If in function x = y2 we switch the x and y we
will get y = x2 . The graph of x = y2 can then be found by switching the x and the y-axis for the graph of
the parabola y = x2 in Fig.1.5a and we get the curve depicted in fig.1.7a. Now we need to consider that
√ √
x = y2 is equivalent to y = ± x. Thus, the upper branch in fig.1.7b corresponds to y = x, whereas the

lower branch corresponds to y = − x. Thus, we need to only draw the upper arm of x = y2 to obtain
√ √
y = x. Similarly, the graph of the function y = 3 x (Fig.1.7c) can be found by a 90o rotation of the
graph of the function y = x3 from Fig.1.6a.
10 P RELIMINARIES

Graphs of rational functions

y= 1 +b slope 1
a
x+a
y y y x
y= 1 1
y= x+a
x a

x b
x x

c
a b
Figure 1.8:
p(x) 1
Here we show examples of graphs of rational functions q(x) . The graph of the function y = x (Fig.1.8a)
1
has a vertical asymptote (x = 0) and a horizontal asymptote (y = 0). The graph of function y = x+a +b
1
can be obtained by a shift of the graph y = x by b units in the y (vertical) direction and by −a units in the
x (horizontal) direction. In this case the vertical asymptote (x value at which function goes to infinity) is
1
x = −a, as at this point the denominator in x+a equals zero. The horizontal asymptote of this graph is
1
y = b, given by lim x+a + b = b. These first two functions are called hyperbolic functions or hyperboles
x→∞
x
for short. Another rational function y = x+a occurs in the classical Michaelis-Menten kinetics. Fig.1.8c
shows the graph of this function. Because for biological applications x and a are always considered
non-negative (x ≥ 0, a > 0), we show the graph in the first quadrant only. We see that independent of the
x
value of the parameter a the horizontal asymptote is always located at y = 1, as lim x+a = 1. The slope
x→∞
x 0
of this function at x = 0 is given by the function derivative f 0 (x) = ( x+a ) = a
(x+a)2
at x = 0, which gives
a slope of f 0 (0) = 1
a.

y= 1 +b
2 2
x+a
1 y y x2
y= x2+ a2
y y= x2 1

b
x x slope=0 x

c
a b
Figure 1.9:
Graphs of similar functions but now involving a second power: y = x12 and y = x2 +a 1
2 , are shown in
1 1
Fig.1.9a,b. We see that function y = x2 has a graph similar to that of y = x but located in the first
and second quadrants, rather than first and third as x2 always has a positive value. Another important
1 2
difference arising for the function y = x2 +a 2 is that it no longer has a vertical asymptote at x = −a (x is

always positive so can never equal −a2 ). Instead, it always reaches a maximum at x = 0. The function
2 n
y = x2x+a2 and its more general form y = xnx+an is a frequently used function in biology, and is called a
x
Hill function. Its graph (Fig.1.9c) has a horizontal asymptote at y = 1 (similar to y = x+a ). However, the
2 x
rate of growth of y = x2x+a2 for small values of x is slower than for y = x+a : the slope of the tangent line
at x = 0 here is 0 rather than 1/a, which can be found from the derivative of this function at x = 0.
1.3 D RAWING OF FUNCTIONS 11

Graphs of exponential and trigoniometric functions

1− sin(x)
λt 1
eλ t λ<0 e λ>0
−π π
1− −1
t t
a b c
Figure 1.10:
Finally in fig.1.10 we show graphs of two other functions that are important in this course eλt and sin(x).
Note that if t grows the function eλt approaches zero if λ < 0 and diverges to infinity if λ > 0. The
function sin(x) oscillates with a period of 2π between −1 and +1.

Drawing graphs

In this course it will often be necessary to draw the graph of f which is a function of a variable x but
also depends on one or more constant valued parameters a, b, c. How should you do this? There are
three complementary approaches possible. For all these three approaches holds that if the graph can have
different possible shapes depending on the parameter values, all different possibilities should be found
and drawn.

The first possibility is to simplify the equation of the function and try to find to which known class of
functions (linear, quadratic, cubic, exponential, etcetera) it belongs. If you know the graphs correspond-
ing to these general functions you can derive how the graph for the function you have been given should
look. For example, the graph of the function y = f (x) + p can be obtained by a vertical shift by p units
of the graph of y = f (x). Similarly the graph of the function y = (x − a)2 can be obtained by a horizontal
shift of a units of the graph of y = x2 . Remember to determine if changes in general graph shape occur
if parameter values change.

The second possibility is to analyse the function and determine whether and where it intersects the x-axis
and y-axis, whether and where it has asymptotes, and whether and where it has maxima.

1. Intersection points with the x-axis you can find by putting f (x) = y = 0 and solving for x.

2. Intersection points with the y-axis you can find by substituting x = 0 in f (x) = y and solving for y.
Note that not all functions have intersections with the x or y-axis.

3. Horizontal asymptotes you can find from the limits y = lim and y = lim .
x→+∞ x→−∞

p(x)
4. For rational functions of the form f (x) = q(x) you can find vertical asymptotes if there are x values for
which q(x) = 0 and f (x) goes to plus or minus infinity. Note that also not all functions have horizontal
and vertical asympotes.

5. Locations (x-value) of maxima and minima of the function f (x) can be found by determining the
derivative f 0 (x) and determining for which x values f 0 (x) = 0. By subsequently filling in these x values
in f (x) you can find the height (y-value) of the maxima and minima.
12 P RELIMINARIES

6. If you need to know whether the function increases or decreases as a function of x for a certain range
of x values (for example between two intersection points with the x-axis), you can determine the value
of the derivative f 0 (x) for these x values. f 0 (x) > 0 means that f (x) locally increases as a function of x,
f 0 (x) < 0 means that f (x) locally decreases as a function of x.

The above mentioned rules are represented graphically in fig.1.11.

y p(x)
y=f(x)=
vertical asymptote q(x*)=0

q(x)
y−intercept y=f(0)
zeros f(x*)=0

horizontal asymptote y=lim f(x)


x−>
x

8
relative extrema df/dx=0

Figure 1.11:

For all the above aspects: intersection points, asymptotes, maxima and minima and slopes hold that they
may depend on parameter values. So, again, remember to draw all possible function shapes.

The third and final possibility is to use your graphical calculator to draw the function by filling in “rea-
sonable” values for the parameters. Remember to always try out multiple combinations of parameter
values to make sure you find all possible graph shapes. Probably, to you this final possibility appears
most easy and attractive. Note however that with this last approach, without having some basic under-
standing of the shape of general functions and how this shape and major shape determinants such as
intersection points, asymptotes and maxima and minima may change due to parameter changes, it is very
hard to determine when you have found all possible function shapes and can stop trying out different
combinations of parameter values!

As we illustrated above, the relation between two variables x and y can be expressed explicitly in terms
of a function y = f (x) that gives us the value of y if we know the value of x. It is also possible that the
relation between x and y is given implicitly as an equation. Such relations are called implicit functions,
and their graphs are called implicit function graphs. One of the most effective methods to plot such
graphs is to try to solve the implicit equation and rewrite it as one or several explicit functions. In some
cases the relations between x and y can be plotted directly. Let us consider two examples:

Example: Draw a graph of the function(s) given by equation: x2 + y2 = C2



Solution: We can either rewrite it as two explicit functions y = ± C2 − x2 and draw the two graphs
given by this equation. Alternatively, we can note that x2 + y2 gives a square of the distance from the
point (0, 0) to the point (x, y), thus equation x2 + y2 = C2 gives the points located at a distance C from
the origin. That is a circle with a radius C with the center at (0, 0) (Fig.1.12).

R 2
Example: Draw a graph of the function(s) given by equation: aR + b R+c − dRN = 0, where R, N ≥ 0 are
1.4 E XERCISES 13

y
(x,y)

y
x
x

Figure 1.12:

the variables and a, b, c, d ≥ 0 are the parameters.

Solution: Let us factor out R in the equation:

R2 R
aR + b − dRN = R(a + b − dN) = 0
R+c R+c
The product of two numbers is zero if one of these numbers is zero, therefore this equation is equivalent
R R
to: R = 0 or a + b R+c − dN = 0. Graphing of R = 0 is trivial. In order to graph a + b R+c − dN = 0
b
R a R
let us rewrite it as dN = a + b R+c , or N = d + R+c
d
. The horizontal asymptote of this graph is: N =
b
a R
lim + R+c
d
= da + b/d = a+b
d . The vertical asymptote occurs if the denominator of the fraction is zero,
R→∞ d
i.e. at R = −c. However, because R ≥ 0 and c > 0 this asymptote will be outside the range of our
function. Additionally note that this function is similar to the graph of Fig.1.8c, but it is shifted upward
by da . Thus the function graph here contains two branches R = 0 and N = da + b/dR R+c that are plotted in
Fig.1.13.

a+b
N d

a
d R

Figure 1.13:

1.4 Exercises

0. Read the chapter

(a) What is the solution of an equation with an unknown variable?


(b) How many equations do you need to find the value of n variables?
(c) What is a derivative?
(d) What is a linear approximation?
(e) What is a limit?
14 P RELIMINARIES

(f) What is a horizontal asymptote?


(g) What is a vertical asymptote?

1. Simplify the below expressions:

(a) (ax − 2by)(3y − 4bx) + 2b(2ax2 + 3y2 ) − 8xyb2


6 5r
(b) r − 30r+5

2. Find limits:
ax+q
(a) lim 2 2
x→∞ c +x
aN 2 +q
(b) lim b 2 2 , d 6= 0
N→∞ N +c +dN

3. Solve the equation for the specified variable:

(a) find r in: 3r + 2 − 5(r + 1) = 6r + 4


(b) find x in: x + 4x = 4
(c) find N in: (b − Nk )N = 0
(d) find N in: (b − d(1 + Nk ))N = 0, d 6= 0; k 6= 0
b
(e) find N in: ( 1+N/h − d)N = 0, b 6= 0;

4. Solve the system of equations for the specified variables:



x − 2y = −5
(a) find x, y in:
2x + y = 10

ax + by = 0
(b) find x, y in:
cx + dy = −b

x(1 − 2x) + xy = 0
(c) find x, y in:
4y − xy = 0
4x − xy − x2 = 0

(d) find x, y in:
9y − 3xy − y2 = 0
b(1 − Rk − d − aN)R = 0

(e) find R, N in: , a, b, d, k, δ 6= 0;
(R − δ)N = 0

5. Find the derivative of f (x):


1
(a) f (x) = x3
(b) f (x) = e−5x ;
(c) y = (4x − x2 )(2x + 3)
x
(d) y = a2 −x2

6. Without plotting the function find the following information about their graph: find the y-intercept
and x intercept; find horizontal or/and vertical asymptotes (if they exist).
an2 −b
(a) function y(n) given by y = n2 +c2
, a, b, c > 0
(b) function N(R) given by N = a (h + R)(1 − KR ),
r
r, a, h, K 6= 0

7. Sketch graphs of the following functions:

(a) y = 3 − 6x
(b) y = x − 3x2
1.4 E XERCISES 15

3x
(c) y = x+a + 4. Find how the shape of the graph for x, y ≥ 0 depends on the value of the
parameter a > 0.

8. (a) Sketch qualitative graphs of the following implicit functions.


(b) Find how special points of these graphs (intercepts, zeros, asymptotes) depend on parameters.
(c) If graph contains several lines find their intersection points.
Note, all parameters represent positive numbers.

(a) x + 3y2 = 0 on the xy plane


(b) y2 + x2 = 9 on the xy plane
(c) xy = 0 on the xy plane
(d) dN(a − P) = 0 on the NP plane
NP
(e) dN + N+a = 0 on the NP plane
cRN
(f) dR(b − R) = R+a on the RN plane
R
(g) bR(1 − − dR) − aNR = 0 on the RN plane
k
(h) aN + P(1 − eP) + bP = 0 on the NP plane

Figure 1.14:

9. Figure 1.14 shows the header of a recent newspaper article. Draw the graph x−x0
a + y−y0
b = 0.
16 P RELIMINARIES
Chapter 2

Vectors, Matrices and Complex Numbers

2.1 Scalars, Vectors and Matrices

Properties that only have a size but no direction are described by a single number, also called scalar.
Examples are the number of individuals in a population N or the concentration level of a chemical
compound in a vessel c.

A vector in 2 dimensions Scaling of vector

y v= a
b( ) y v= a
b ( )
b
0.5b )
( 0.5a
b
w=0.5v=
0.5b

a x 0.5a a x

Adding of 2 vectors Rotation of a vector


y u=v+w= a+c
b+d ( ) y v= a
( )
b
b b
w=Tv= c
( )
w= c
d( ) d
( )
v= a
b T c
d

c a x a x
d

Figure 2.1: Vector in a 2 dimensional plane, scaling of a 2D vector, adding of two 2D vectors and rotation
of a 2D vector.

If, in contrast, some property has both a size and a direction, we use a vector to describe it (Figure 2.1).
Vectors can be used to describe the forces acting on an object, or the speed and direction with which an
object moves. Mathematically, vectors are written as a row or column of numbers between brackets, and
are thus referred to as either a row or column vector. For example, in a two dimensional plane in which
the force acting on an object has an x-component Vx = 2 and an y-component Vy = 1 this force can be
18 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

represented as:  
~V = 2 ~V =

or 2 1
1

The length of the force vector is then given by |V | = 22 + 12 , whereas the direction of the force vector
is 2 steps (in a certain unit) in the positive x-direction (to the right) and 1 step in the positive y-direction
(upward).

The simplest operation that can be performed on a vector is it’s multiplication by a scalar, as the word
scalar implies, this simply results in the scaling of the size of the vector, without changing its direction
(figure 2.1):      
~ 2 0.5 ∗ 2 1
0.5V = 0.5 = =
1 0.5 ∗ 1 0.5
an example would be a car that keeps driving in the same direction but halves it’s speed.

Another important operation is the adding of two or more vectors, for example to determine the net
resultant force from the sum of all forces acting on an object. Vector addition is achieved by simply
adding up the corresponding elements of the different vectors (Figure 2.1). Thus, addition can only be
performed on vectors that are of the same size (same number of elements):
       
~ = 2 + 1 = 2+1 = 3
~V + W
1 3 1+3 4

A more complex operation is the rotation of a vector. Such a rotation can be obtained by multiplying the
vector by a so called matrix: A~V = W
~ , where ~V is the original vector, W
~ is the new resulting vector, and
A is the matrix performing the rotation (Figure 2.1).

Before explaining this in more detail, let us first introduce the concept of a matrix. Mathematically
speaking, matrices are written as a block of n rows and m columns of numbers, all between brackets:
 
1 4 5
A= (2.1)
2 6 10
This particular matrix A has two rows and three columns, so we say it has a size of 2 × 3. Matrices can
be used to store and represent data sets. As an example we can summarize an experiment in which the
expression of a large set of genes is measured over a range of different conditions. By using the rows to
represent the different genes and the columns to represent the different conditions, each matrix element
reflects the expression level of a single gene under a particular condition.

As for vectors, the simplest operation that can be performed on a matrix is the multiplication by a scalar.
This is done by multiplying each individual element of the matrix with the scalar. For a general 2 × 2
matrix this can be written as:    
a b λa λb
λ =
c d λc λd
a simple example would be the rescaling of experimentally measured fluorescence levels by a factor
translating fluorescence level into gene expression level to find the levels of gene expression for all
conditions and genes.

Like vectors, two matrices A and B can be added up into a new matrix C only if they are of the same
size (both the number of rows and the number of columns should be equal). Matrix addition can then be
performed by adding up the corresponding elements of the two matrices:
       
1 4 5 2 1 4 1+2 4+1 5+1 3 5 6
+ = = (2.2)
2 6 10 1 3 5 2 + 1 6 + 3 10 + 5 3 9 15
2.1 S CALARS , V ECTORS AND M ATRICES 19

For two general 2 × 2 matrices this can be written as:


     
a b x y a+x b+y
+ = (2.3)
c d z w c+z d +w

For the elements in a matrix C = A + B this can be written as:

ci j = ai j + bi j (2.4)

a simple example would be if a matrix represents what a shop has in storage, with in the rows different
items and in the columns different sizes, than the sum of two such matrices is what two shops have in
storage.

Finally, one can multiply a matrix A with a matrix B to obtain a new matrix C if the number of columns
in the first matrix is equal to the number of rows in the second matrix. Matrix multiplication is defined
as the products of the rows of the first matrix with the columns of the second matrix. Thus, to find the
element in row i and column j of the final matrix we need to multiply the ith row of the first matrix by
the jth column of the second matrix:
p
ci j = ∑ aik bk j (2.5)
k=1
For a product of two 2 × 2 matrices this gives:
    
a b x y ax + bz ay + bw
= (2.6)
c d z w cx + dz cy + dw

From this it follows that multiplication of a matrix by a column vector (simply a matrix with only one
column) is given by:     
a b vx avx + bvy
= (2.7)
c d vy cvx + dvy
 
a b
Note that in equations 2.6 and 2.7 multiplication with the first matrix A = produces a trans-
c d
formation of the second matrix or vector it is multiplied with. I.e. the first matrix is the transformation
matrix, the second matrix or vector is the one being transformed.

With this knowledge we can now return to the problem of rotating a vector by multiplying it with a ma-
trix. It turns out that to rotate a vector in a two-dimensional plane with an angle θ in a counterclockwise
direction we need to multiply this vector with a 2 × 2 transformation matrix:
 
cosθ −sinθ
A=
sinθ cosθ
Applying this matrix we then obtain:
    
cosθ −sinθ vx cosθvx − sinθvy
A~v = =
sinθ cosθ vy sinθvx + cosθvy

To illustrate how this matrix transformation actually works, we consider a simple example in which we
can compute the new, rotated vector both by using the method of matrix transformation and by using
simple trigonometry. We consider the vector vx = 1, vy = 0, which lies on the x-axis and has a length
l = 1, and rotate it counterclockwise by an angle θ (see fig.2.2). From basic trigonometry we know that
cosθ = wx /l = wx and sinθ = wy /l = wy and hence that wx = cosθ and wy = sinθ. Now let us check
whether we find the same answer by applying the above transformation matrix on the vector. Applying
A~v now gives us wx = cosθvx − sinθvy = cosθ, wy = sinθvx + cosθvy = sinθ (vy = 0). So indeed we obtain
20 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

1/2
y L=|v|=(1*1+0*0) =1
w cos(Phi)= w x /L=wx
wy
sin(Phi)= w y /L=w y

Phi
wx v x

Figure 2.2: Rotation of a simple vector with length 1 and oriented parallel to the x-axis.
Other matrix transformations of vectors:
Complex scaling of vector

y v= a
b ( )
b
w=Tv= 0.5 0
0 (
0.25
v= 0.5a
0.25b ) ( )
0.25b

0.5a a x

Shearing of vector parallel to x-axis Shearing of vector parallel to y-axis

y
v= a
b ( ) y
0.2a+b
( )
v= a
b

b
w=Tv= 1
0 ( 1 ) (
0.2 v= a+0.2b
b ) b w=Tv= 1 ( 0
0.2 1 ) v= ( 0.2a+b
a
)
a+0.2b a x
a x

Figure 2.3: Scaling and shearing matrix transformations of vectors.

the same result. For more complex situations, in which the initial vector is not aligned to one of the axis
and does not have a length 1, it is not easy to derive wx and wy using only trigonometry, but the method
of A~v = ~w always works.

Rotation of a vector by applying a rotation matrix, is only one of a series of possible transformations that
can be applied on vectors by specific matrices. If we want to scale a vector in a different amount in the x
and y direction this can also be performed by a transformation matrix (see figure 2.3):
    
wx sx 0 vx
=
wy 0 sy vy

Similarly, a shear matrix can be applied to apply shearing of a vector, either parallel to the x-axis see
figure 2.3):
    
wx 1 k vx
=
wy 0 1 vy
this results in wx = vx + kvy and wy = vy .
2.2 M ATRICES AND SYSTEMS OF EQUATIONS 21

or parallel to the y-axis see figure 2.3):


    
wx 1 0 vx
=
wy k 1 vy
which results in wx = vx and wy = vy + kvx .

Figure 2.4: On the left the original pictures from the 1917 book “On Growth and Form” by D’Arcy
Wentworth Thompson. On the right pictures from a computer program of the School of Mathematics
and Statistics of the University of StAndrews in Scotland performing similar shape transformations.
Transformatin matrices used apply rotation, scaling and shearing transformations.

Note that as matrices can be multiplied with one another to obtain a new matrix, multiplication of rota-
tion, scaling and shearing matrices can result in a single matrix that performsa complex
 transformation.
x
Also note that a point (x,y) on an object can be considered as a vector ~v = . Adding these two
y
things up, we can apply complex transformation matrices to objects (basically a collection of points) to
change them into objects with different orientations, shapes and sizes. This can for example be applied to
simulate or deduce the changes in shape and size that occur during development, evolution or growth in
animals and plants. Indeed, the famous mathematician and biologist D’Arcy Wentworth Thompson used
transformation matrices to show how you could go from the shape of one fish species to that of another
fish species, or from the shape of a human skull to the shape of a chimpanzee skull and called this the
theory of transformation which he described in his 1917 book “On Growth and Form” (see figure 2.4).

Note that matrix multiplications are not commutative, i.e. A × B 6= B × A. This means that if the same
transformations are applied in a different order, a different outcome will be produced.

2.2 Matrices and systems of equations

In this course we will often use matrices as an alternative way to write down systems of linear equations.


x − 2y = −5
(2.8)
2x + y = 10
22 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

we can write the coefficients in front of x and y in the left hand side as a square matrix:
 
1 −2
A= .
2 1

We also have two numbers in the right hand side which we can write as a column vector:
 
~V = −5 .
10

Now if we write x and y as a column vector:


 
~X = x
.
y

Thus we can represent the system of (2.8) using matrix multiplication (2.7) as:
    
~ ~ 1 −2 x −5
AX = V ; or = (2.9)
2 1 y 10

What is the solution of this system? From x − 2y = −5 we obtain x = 2y − 5. Filling this in in 2x + y = 10


gives us 2(2y − 5) + y = 10, 4y − 10 + y = 10, 5y = 20 so y = 4. Filling this back in x = 2y − 5 gives
x = 2 × 4 − 5 = 8 − 5 = 3. So the solution is:
   
x 3
=
y 4

( ac bd ) ,vectors v= ( c ) ( bd )
Matrix: A= a
and w=

y
surface parallellepipidum?

c+d (a+b)*(c+d)-A-B-C-D-E-F=
A F ac+ad+bc+bd
d -bc-bc
E
c C -0.5bd-0.5ac
B -0.5bd-0.5ac=
D ac+ad+bc+bd
b a a+b x -2bc-bd-ac=
ad-bc

Figure 2.5: The determinant of a matrix can be used to compute the surface of the parallellepipidum
spanned by the vectors constituting this matrix

An important property of a square matrix such as the one above is its so-called determinant. In this
course we will onlyuse determinants
 of 2 × 2 matrices which are defined as follows:
a b
For the matrix A = the determinant is
c d

a b
det
= ad − cb (2.10)
c d

If we consider the different columns of the matrix as vectors, then these vectors span a parallellepipidum,
and the determinant of the matrix gives us the volume occupied by this parallelllepipidum (see figure
2.5). However, originally determinants were invented to study whether a system of linear equations can
be solved. It can be shown that only for a determinant unequal to zero this is possible. To see why this
is the case consider the following general linear system:

ax + by = p
(2.11)
cx + dy = q
2.3 E IGENVALUES AND EIGENVECTORS 23

Let’s first solve the first equation: ax + by = p so ax = p − by so x = (p/a) − (b/a)y. Let’s use this
solution to solve the second equation: cx + dy = q. Together with x = (p/a) − (b/a)y this gives us
(cp/a) − (cb/a)y + dy = q and hence (d − (cb/a))y = q − (cp/a) or ((da − cb)/a)y = ((qa − cp)/a).
This can be solved into y = (qa − cp)/(da − cb) and this has a finite  only if da − cb is not equal
 solution
a b
to zero. We already know that the matrix for this system is A = and that it has a determinant
c d
ad − cb. So indeed if the determinant equals zero da − cb equals zero and no solution can be found. Note
that we can also determine the determinant of larger square matrices, but their treatment is outside the
scope of this course.

2.3 Eigenvalues and eigenvectors

We have learned above that using matrices we can transform vectors, changing both their size and their
direction. It turns out that for each particular matrix there exists a set of special vectors called eigenvec-
tors that only change their size but not their direction when the matrix is applied to them. The factor
by which the eigenvector changes its size when the matrix is applied to it is called the corresponding
eigenvalue.

Formally, we can write this as follows:

Av = λv (2.12)
which says that for a certain vector v application of the transformation matrix A results only in the
scaling of this vector by an amount λ. Thus, v is an eigenvector and λ the corresponding eigenvalue of
transformation matrix A.

Note that eigenvectors are not unique in the sense that you can always multiply them by an arbitrary
constant k to get another eigenvector:

kAv = kλv or A(kv) = λ(kv) (2.13)

therefore, we can say that kv is also an eigen vector of (2.12) corresponding to eigen value λ.
y
v2

(
A= 1 2
2 1 )
eigenvalues: 3, -1

eigenvectors
( 11 ) (-11 )
v0= 0
1( ) ( )
v1= 2
1
v2= 4
5 ( ) ( )
v3= 12
13

eigenvector 1 v1

v0 x
eigenvector 2

Figure 2.6: Repeated transformation of a vector v0 by a matrix A produces a vector with a direction closer
and closer to the eigenvector of matrix A with the largest eigenvalue.

What is the use of knowing eigenvalues and eigenvectors of a matrix and hence a system of equations? It
can be shown that eigenvectors give the principal directions of change imposed by a matrix, whereas the
eigenvalues give the amount of change in each of these directions. Thus, if we know the eigenvector with
24 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

the largest eigenvalue, we can to a large extent predict the effect of a matrix and hence the behaviour of
the system: it will be rotated into the direction of this dominant eigenvector (see Figure 2.6).

Finding eigen values and eigen vectors is one of the most important problems in applied mathematics. It
arises in many biological applications, such as population dynamics, biostatistics, bioinformatics, image
processing and many others. In our course we will apply it for the solution of systems of differential
equations, which we will consider in Chapter 4.

Let us consider how to solve the eigenvalue problem for a 2x2 matrix (i.e. find the eigenvalues λ and
eigenvectors v):
    
a b vx vx
Av = =λ . (2.14)
c d vy vy

We can rewrite this as a system of two equations with three unknowns λ, vx , vy :



avx + bvy = λvx
(2.15)
cvx + dvy = λvy

This can be further rewritten as:


     
(a − λ)vx + bvy = 0 a−λ b vx 0
or in matrix f orm = . (2.16)
cvx + (d − λ)vy = 0 c d −λ vy 0
This system always has a solution vx = vy = 0. However, eigenvectors are defined to be nonzero vectors,
therefore this solution doesn not correspond to an eigenvector. In order to find non-zero solutions let us
multiply the first equation by d − λ, the second equation by b and subtract them.
Multiplication gives: 
(d − λ)[(a − λ)vx + bvy ] = 0
(2.17)
b[cvx + (d − λ)vy ] = 0
or 
(d − λ)(a − λ)vx + (d − λ)bvy = 0
(2.18)
bcvx + b(d − λ)vy = 0
Substraction then gives:

(d − λ)(a − λ)vx − bcvx + (d − λ)bvy − b(d − λ)vy = 0

or
[(d − λ)(a − λ) − bc]vx = 0 (2.19)
given that v 6= 0 we get:

(d − λ)(a − λ) − bc = λ2 − (a + d)λ + (ad − cb) = 0 (2.20)

This is a quadratic equation with unknown λ for which the two possible solutions λ1 and λ2 can be found
using the ’abc’ formula. In general, for a n × n matrix there are maximal n solutions for λ. Equation
(2.20) is called the characteristic equation, and will often be used in this course.

This characteristic equation can also be written in matrix format. To derive this formulation recall that
the determinant of a matrix is given by the following equation:

a b
det = ad − cb.
c d
2.3 E IGENVALUES AND EIGENVECTORS 25

 
a−λ b
Similarly the determinant of matrix is:
c d −λ

a−λ b
det = (a − λ)(d − λ) − bc (2.21)
c d −λ

which coincides with the left hand side of characteristic equation (2.20) and thus the characteristic equa-
tion can be rewritten as:
a−λ b
det =0 (2.22)
c d −λ

Example: Let us use this approach to find the eigen values of the following matrix:
 
1 2
2 1

To do this we write the following characteristic equation:



1−λ 2
Det = (1 − λ)(1 − λ) − 2 ∗ 2
2 1−λ
= 1 − λ − λ + λ2 − 4 = λ2 − 2λ − 3 = 0

From the ’abc’ formula it then follows that:


√ √
2 ± 4 + 12 2 ± 16
λ1,2 = =
2 2
this gives us λ1 = 3 and λ2 = −1. As a next step we have to find the eigenvectors belonging to these two
eigenvalues. We can do this by substituting the found eigenvalues into the original equations and solving
the equations for vx and vy . For the eigenvector corresponding to the eigenvalue λ1 = 3 we obtain:

  
(1 − 3)vx + 2vy = 0 −2vx + 2vy = 0 −2vx = −2vy
or or (2.23)
2vx + (1 − 3)vy = 0 2vx − 2vy = 0 2vx = 2vy

The two   give us the same solution: vx = vy . This means that if vy = 1, then vx = 1, and thus
equations
1
that v = is an eigenvector corresponding to the eigenvalue 3. However, we can use any other
1
value for vx and hence vy as long as vx = vy is satisfied. In general any vx = k, and vy = k will give us an
eigenvector corresponding to the eigenvalue 3. We can write this as:
   
vx 1
=k (2.24)
vy 1

where k is an arbitrary number. Formula (2.24) thus gives us all possible solutions of eq.(2.23). It also
illustrates a general property of eigenvectors which we have already proven in (2.13), that if we multiply
an eigen vector by an arbitrary number k will get another eigen vector of our matrix.

Similarly we can find the eigenvector corresponding to the other eigen value λ2 = −1:

1. Substitution:
  
(1 − (−1))vx + 2vy = 0 2vx + 2vy = 0 2vx = −2vy
or or (2.25)
2vx + (1 − (−1))vy = 0 2vx + 2vy = 0 2vx = −2vy
26 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

2. Relation between vx and vy :


vx = −vy
3. Eigenvector: use e.g. vy = 1, thus vx = −1
 
−1
v=
1
 
−1
The general form then is v = k , where k is an arbitrary number.
1

Note, that in both cases in order to find the eigenvectors we could have used only the first equation(see
equations (2.23) and (2.25)). The second equation in both cases did not provide us with any new infor-
mation. We can use this property for a fast and easy method to find eigenvectors, which is discussed
below.

Express method for finding eigen vectors

Let us derive a formula for finding the eigenvectors of a general system (2.15). We assume that we have
found eigenvalues λ1 and λ2 from the characteristic equation (2.22). To find the corresponding eigen
vectors we need to substitute the found eigen values into the matrix and solve the following system of
linear equations (2.16): 
(a − λ1 )vx + bvy = 0
(2.26)
cvx + (d − λ1 )vy = 0
It is easy to check that if we use for vx and vy the values vx = −b and vy = a − λ1 it gives the solution of
the first equation:

(a − λ1 )vx + bvy = (a − λ1 )(−b) + b(a − λ1 ) = −b(a − λ1 ) + b(a − λ1 ) = 0

If we substitute these expressions into the second equation we get:

cvx + (d − λ1 )vy = −cb + (d − λ1 )(a − λ1 ) = 0

To prove that this expression is also zero, note that (d − λ1 )(a − λ1 ) − cb is zero in accordance with the
characteristic equation (2.20). Therefore vx = −b and vy = a − λ1 give a solution of (2.26) which is an
eigenvector corresponding to the eigen value λ1 . Similarly we find the eigenvector corresponding to the
the eigenvalue λ2 .

However, this approach does not work if in (2.16) both b = 0 and a − λ = 0. In this case we can use the
second equation cvx + (d − λ1 )vy = 0 and find an eigenvector as vx = d − λ1 and vy = −c. Indeed:

cvx + (d − λ1 )vy = c(d − λ1 ) + (d − λ1 )(−c) = 0

As in the previous case it is easy to show that this vector satisfies the other (first) equation as: (a −
λ1 )vx + bvy = (a − λ1 )(d − λ1 ) + b(−c) = 0 due to (2.26).

Summarizing, the final formulas are:


       
v1x −b v2x −b
v1 = = v2 = = (2.27)
v1y a − λ1 v2y a − λ2
2.4 C OMPLEX NUMBERS 27

or        
v1x d − λ1 v2x d − λ2
v1 = = v2 = = (2.28)
v1y −c v2y −c

a b
where a, b, c, d are the elements of the matrix A = . Either (2.27) or (2.28) can be used to find
c d
eigenvectors. (Both answers will be valid.) If, however, one of the formulas gives a zero eigen vector,
we should use the other one to obtain a non-zero vector.
 
1 2
Example: Let us apply these formulas for the system with matrix A = and eigen values
2 1
λ1 = 3; λ2 = −1. The eigen vectors can be found from (2.27) as:
           
v1x −2 −2 v2x −2 −2
λ1 = 3; = = λ2 = −1; = )=
v1y 1 − (3) −2 v2y 1 − (−1) 2
(2.29)

and from (2.28) as:


           
v1x 1−3 −2 v2x 1 − (−1) 2
λ1 = 3; = = λ2 = −1; = )= (2.30)
v1y −2) −2 v2y −2 −2

We see that the vectors we found differ from the vectors we found earlier, butit is easy
 tosee that
 they are
−2 1
equivalent. For example, if we multiply the first vector by − 21 we find − 12 = , which is
−2 1
the same vector as we found earlier in (2.24). We also see that formulas (2.29) and (2.30) give equivalent
results. Indeed, the first
 vectors
 obtained
 form (2.29) and (2.30) are exactly the same. For the second
−2 2
vectors note that: −1 = .
2 −2

2.4 Complex numbers

Consider the general quadratic equation:

λ2 + Bλ +C = 0 (2.31)

The roots of (2.31) are given by the well known ’abc’ formula:
√ √
−B ± B2 − 4C −B ± D
λ12 = = (2.32)
2 2
where
D = B2 − 4C (2.33)
What happens with this equation if D < 0? Does the equation still have roots in this case?

So-called complex numbers have been invented exactly to solve this particular problem and still allow
for a solution if D < 0. Let us define the basic complex number i as:

i = −1

or equivalently
i2 = −1
28 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

A general complex number z can then be written as z = α + iβ, where α is called the real part (Reaz)
and iβ is called the imaginary part (Imz) of the complex number z. Complex numbers are often plotted
on a complex plane, a graph in which the x-axis is used for the real part and the y-axis is used for the
imaginary
 part of the complex number (see Figure 2.7). Note how this is similar to the depiction of a
x
vector , when x and y are real valued numbers in real plane. Indeed it turns out that scaling, adding
y
and multiplication of complex numbers follows the same rules as for vectors.
The complex plane

imaginary part
complex number z

real part

Figure 2.7: Depiction of complex number on complex plane

To see how this works, consider the following equation:


λ2 = −1 (2.34)
Given that i2 = −1 we can now claim that (2.34) has a solution and denote it as:
λ12 = ±i (2.35)
Here i is the basic complex number which is similar to 0 10 for real numbers. Using it we can denote
solutions of other similar equations. For example if
√ √ √
λ2 = −4, λ = −1 × 4 = −1 4 = i × (±2) = ±2i.
Similarly, the equation λ2 = −a2 , has solutions λ = ±ai. Although we call ai a complex number, it is
quite different from usual real numbers. Using complex numbers ai we cannot count how many books
we have in the library, for example. The only meaning of i is that i2 = −1, and ai is just a designation of
a root of the equation λ2 = −a2 .
√ √
Now we can solve equation (2.31) for the case D < 0. If D < 0, then D = i −D and

−B ± i −D
λ12 = (2.36)
2

Example. Solve the equation λ2 + 2λ + 10 = 0

Solution. √ √
−2 ± 4 − 4 ∗ 10 −2 ± −36 −2 ± 6i
λ12 = = = , (2.37)
2 2 2
or λ1 = −1 + 3i, λ2 = −1 − 3i. We see that solution of this equation λ1,2 is complex, and has a real part
Rez ’-1’, which is the same for λ1 and λ2 and an imaginary part Imz ±3i which has opposite signs for λ1
and λ2 .

As stated above, scaling, adding and multiplication of complex numbers follows the same rules as for
vectors. The only extra thing which we need to remember, is that i2 = −1. To add two complex numbers
we need to add their respective real and imaginary parts. For example:
z1 = 3 + 10i, z2 = −5 + 4i, z1 + z2 = (3 + 10i) + (−5 + 4i) = 3 + 10i + −5 + 4i = −2 + 14i.
2.4 C OMPLEX NUMBERS 29

Similarly, multiplication by a real number results in the multiplication of both the real and imaginary
part by this number:
z1 = 3 + 10i; 10z1 = 10(3 + 10i) = 30 + 100i.

Multiplication of two complex numbers is just an exercise in multiplication of two expressions, i.e. the
real and imaginary part of number 1 should both be multiplied by both the real and imaginary part of
number 2: z1 = 3 + 10i, z2 = −5 + 4i; z1 z2 = (3 + 10i)(−5 + 4i) = 3(−5) + 3 ∗ 4i + 10i(−5) + 10i4i =
−15 + 12i − 50i + 40i2 = −15 − 38i − 40 (as i2 = −1) = −55 − 38i.

Similarly

(z1 )2 = (3 + 10i)2 = 32 + 2 ∗ 3 ∗ 10i + (10i)2 = 9 + 60i + 100i2 = 9 + 60i − 100 = −91 + 60i.

Now we can check that λ1 = −1 + 3i is indeed a solution of the equation in example (2.37). Indeed:
λ2 + 2λ + 10 = (−1 + 3i)2 + 2(−1 + 3i) + 10 = (−1)2 + 2(−1)3i + (3i)2 − 2 + 6i + 10 = 1 − 6i − 9 − 2 +
6i + 10 = (1 − 9 − 2 + 10) − 6i + 6i = 0 − 0i = 0.

One more definition. The number z2 = a − ib is called the complex conjugate to the number z1 = a + ib
and is denoted as z̄1 = z2 = a − ib. Complex conjugate numbers have the same real parts, but their
imaginary parts have opposite signs.

Roots of a quadratic equation with negative discriminant D < 0 are always complex conjugates to each
other. This follows from formula (2.36)
√ √
−B + i −D −B − i −D
λ1 = λ2 = (2.38)
2 2
hence: √ √
−B −D −D
Reλ1 = Reλ2 = ; Imλ1 = ; Imλ2 = − . (2.39)
2 2 2


Finally, consider two more basic operations. If z = a + ib, then, |z| = a2 + b2 is called the absolute
value, or modulus of z (similar to the length of a real valued vector). Note, that |z|2 = zz̄, as (a + ib)(a −
ib) = a2 − (ib)2 = a2 + b2 .

We use this trick to perform the division of two complex numbers

z1 z1 z¯2
=
z2 z2 z¯2

So, to divide two complex numbers, we multiply the numerator and the denominator by a number which
is the complex conjugate to the denominator, and we get the answer in the usual form.

Example

1 + 3i 1 + 3i 1 + 4i (1 + 3i)(1 + 4i) 1 + 3i + 4i + 12i2 −11 + 7i −11 7


= = 2 2
= = = + i
1 − 4i 1 − 4i 1 + 4i 1 +4 17 17 17 17

Determinants, eigenvalues, eigenvectors and complex numbers will all come in handy later in this course
when trying to determine the behaviour of systems of differential equations.
30 V ECTORS , M ATRICES AND C OMPLEX N UMBERS

2.5 Exercises
0. Read the chapter

(a) Explain the difference between a scalar, vector and matrix

(b) Can you add up all vectors or matrices?

(c) Can you multiply all matrices?

(d) What is the relation between a matrix and a linear system of equations?

(e) What is an eigenvalue?

(f) What is an eigenvector?

(g) What is a complex number?

(h) What is a complex conjugate?

(i) What is the modular of a complex number?

Figure 2.8: Picture of a stylized leaf with coordinates.

1. Figure 2.8 shows a drawing of a stylized leaf. The coordinates of the points determining the shape
of the leaf are indicated. Assume that the growth of the leaf can be described by the multiplication
of two transformation matrices A and B on each of the points of the leaf:

Xnew = B × A × Xold

where  
1.5 0
A=
0 2
and  
cosθ −sinθ
B=
sinθ cosθ
and θ = 30degrees

(a) interpret the meaning of A and B

(b) compute C = B × A
2.5 E XERCISES 31

(c) will A × B give the same matrix C?

(d) draw the new leaf that arises after the growth described by C = B × A.

2. Find all roots of the given equations

(a) x2 + 4x + 5 = 0

(b) x2 − 5x + 6 = 0

3. Write the following linear systems in a matrix form A~X = ~V . Find the determinant of matrix A.

2x − 4y = 3
(a)
x+y = 1

ax + by = 0
(b)
cx + dy = −b
4. Find eigen values and eigen vectors of the following matrices:
 
−2 1
(a)
1 −2
 
1 4
(b)
1 1
 
−1 5
(c)
−1 3
32 V ECTORS , M ATRICES AND C OMPLEX N UMBERS
Chapter 3

Differential equations of one variable

With this Chapter we come to the main subject of this course: differential equations.

A differential equation describes the rate of change of a variable:

dx
= ....
dt
for example the rate at which the numbers of individuals in a population changes. Remember that the d
in dx and dy stands for delta or a change in .... Indeed, the amount of change of x, dx divided over a time
interval dt in which it occurs represents the rate of change of x.

In contrast, the solution of a differential equation -which can be obtained by solving/integrating the
differential equation- describes the amount or size of a variable as a function of time:

x(t) = ....

for example the number of individuals in a population over time. It is crucial that you keep this distinction
between differential equation and its solution in mind.

Differential equations are very often used in modeling biological, medical, physical, and chemical pro-
cesses. The great advantage of working with differential equations is that it is much easier to write
equations for how processes produce change in a variable, than writing an equation for how these pro-
cesses influence the size of the variable as a function of time. As an example, it is much easier to see how
birth and death processes translate into a rate of change of population size than to see how they translate
into how population size depends on time.

3.1 Simple differential equations and their solutions

We start with the simple differential equation for the position of a car moving at constant speed a:

dx
=a (3.1)
dt
Eq. 3.1 describes the rate of change in position of the car. However, we are interested in the position of
the car as a function of time x(t). What would the function x(t) look like? If x increases with a constant
34 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

Figure 3.1: Graphical representations of the solutions of equations 3.1 (a), 3.4 (b) 3.6 (c) and 3.8. The
intercept with the y-axis b depicts the initial value for x or N.

amount of a per time unit, x as a function of time can be drawn as a straight line with slope a (See Figure
3.1a). As we do not know the value of x at t = 0 we do not know the height of the intersection point
with the y-axis, and therefore refer to it as b. Thus we can now write down an equation for the solution
(integral) of differential equation 3.1 as:
x(t) = at + b (3.2)
where b = x(0) is the unknown intial position of the car. We call Eq. 3.2 the general solution of Eq. 3.1,
it is general because it can be used for any value of b = x(0). We can simply check this by differentiating
x(t) to t:
dx(t) d(ax + b)
= =a
dt dt
Now let us assume that we know that a particular car started at position x(0) = 10 from a certain reference
point (i.e. the point of which we say that x = 0). In this case, we need to solve Eq. 3.1, while taking
x(0) = 10 into account.Thus, we need to obtain a particular rather than a general solution. In mathematics
this is called the solution of the initial value problem of Eq. 3.1 and x(0) = 10. This solution can be
written as:
x(t) = at + 10 (3.3)

Next let us consider the slightly more complicated example of a car moving with a constant accelleration
a, an example you may remember from high school physics:
dx
= at (3.4)
dt
Simple integration of the function at to time gives us the general solution:
1
x(t) = at 2 + b
2
again with b = x(0) being the initial position of the car. In Figure 3.1b we see that the position of the car
as a function of time changes in a parabolic fashion, with the starting height of the parabola determined
by the initial position. We can again check this solution by differentiating it to t:
dx(t) d((1/2)at 2 + b) 1
= = 2 at = at
dt dt 2
3.1 S IMPLE DIFFERENTIAL EQUATIONS AND THEIR SOLUTIONS 35

If we know that the car started out at position x(0) = 25 we can write a particular solution of this initial
value problem as:
1
x(t) = at 2 + 25
2

Now let us move to the simplest possible biological example. If N is the population size (number of
individuals) of a certain species we can write for the rate of change of the population size is:

dN/dt = births − deaths (3.5)

We now assume that the birth and the death terms are proportional to N. This means that there is a
constant rate or probability per individual to give birth to a new individual or to die, and that these proba-
bilities do not depend on the number of individuals currently present in the popultion. As a consequence
each term in equation (3.5) will be proportional to N and we get the following famous ’Malthus’ equation
for population dynamics:
dN/dt = (b − d)N = kN (3.6)
where b and d are the rate constants for the birth and death processes which can be summarized into a
net growth rate per individual of k. It can be easily seen that k > 0 if b > d, and k < 0, if b < d. Solving
this type of differential equation, in which the right hand side of the equation depends on the variable
itself (N) rather than being constant (Eq 3.1 or only depending on time (Eq. 3.4) is less easy. We will not
find the solution here -for interested students this can be found in the last section of this Chapter. Instead
we will give the solution and proof that it is correct. The general solution of Eq. 3.6 is:

N(t) = Aekt (3.7)

with A = N(0) the population size at t = 0. Now let us proof that this is indeed so. First let us differentiate
N(t) to t:
dN(t) d(Aekt )
= = kAekt
dt dt
Now we need to fill in N(t) and dN(t) dN
dt into dt = kN (note that this is different from above, in which we
only needed to differentiate the solution to t and where then finished):

kAekt = kN(t) = kAekt

So our solution is correct. Note that N(t) = Aekt is the general solution, whereas N(t) = N0 ekt is a specific
solution to the initial value problem N(0) = N0 . Figure 3.1c shows a graph of this exponential population
growth as a function of time. To have a measure of how fast the population grows, the characteristic
time is defined as the time needed to increase the population size e ∼= 2.73-fold. From eN(0) = N(0)ekt
we find that τ = 1k is the characteristic time for exponential growth at rate k.

However, even proofing that a differential equation has a given solution becomes difficult for more com-
plicated equations. As an example, if we have a population in which new individuals arise through
immigration from elsewhere rather than birth we can write the following immigration-death equation:
dN
= k − dN (3.8)
dt
The analytical solution of this equation has the following form:
k
N(t) = (1 − e−dt ) + N(0)e−dt
d
with A = k − dN(0), or N(0) = dk − Ad . In one of the exercises you will check that this solution is correct
for the situation when N(0) = 0, which makes the equation somewhat simpler. Figure 3.1d shows a graph
36 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

of this type of population growth as a function of time. We see that over time N(t) approaches (k/d),
which can easily be seen from the equation as both e−dt terms approache to zero as t increases. Thus
(k/d) is the equilibrium size that the population will reach. For this type of system the characteristic
time is defined as the time needed to change the difference between the current population size and the
equilibrium population size by a factor e. This characteristic time is τ = d1 and can be found from:
e(N(0) − k/d) = N(t) − k/d = (N(0) − k/d)e−dt so e = e−dt .

We have shown here that for very simple, linear differential equations analytical solutions can be ob-
tained. We have also shown what these solutions tell us about the dynamical behaviour of a system.
However, for less simple differential equations it is often very difficult or even impossible to derive an
analytical solution. We will next explain qualitative techniques to determine the behaviour of a system
without having to find the analytical solution.

3.2 Qualitative analysis of differential equations

Phase portrait

Figure 3.2: Three solutions of equation dN


dt = 4N together with tangent lines for these solutions at N = N ∗ .

Consider again the simplest differential equation for population growth dN dt = kN. In the previous section
kt
we learned that the general solution is N(t) = Ae , with k = b − d and A = N(0). Let us assume that for
a particular biological species we know that k = 4. Let us also assume that for a particular population of
that species we know that A = N(0) = 10. Figure 3.2, line a shows the graph of the particular solution
corresponding to this initial value problem. Lines b and c show two different solutions, corresponding to
different populations with different initial sizes N(0) (N(0) = 5 line b, N(0) = 1 line c).

Of course, one might be interested in quantitative measures: the exact size of a population at a particular
moment in time. However, more often we are interested in whether a population is stable, growing or
declining, as this is the most crucial information. In other words, most often we are not interested in the
exact height (y-value) of the graphs in Figure 3.2, but in their slopes. Now we can make an important
observation. In Figure 3.2 we drew a line at N = N ∗ , with which lines a, b and c intersect. As the different
populations (different lines) started at different initial sizes N(0) (different intercepts with the y-axis), it
takes them a different amount of time t to reach the level N = N ∗ . However, at this level all three lines
have the same slope. Put differently, the slope, which is the rate of change of N, is independent of time
and depends only on the size of N.

We can use this dependence on N only to summarize the dynamical behaviour of all possible solutions
of dN
dt = 4N using a a qualitative, graphical description:
3.2 Q UALITATIVE ANALYSIS OF DIFFERENTIAL EQUATIONS 37

N’= 4 8 12 16 20
1 2 3 4 5
N
Here we draw along the N axis the slope or rate of change of N(t) corresponding to the local value of
N. However, we can generalize this approach even further. Although the precise value of the population
growth rate can be interesting, what is really important is the sign of the growth rate: whether it is positive
and the population is growing or whether it is negative and the population is shrinking. Thus, the most
important qualitative aspect is whether the slope or rate of change of N(t) is positive or negative. In
this particular case we see that it is positive for all values of N. We can represent this in the following
qualitative way:

N
i.e. we display the positive slope, i.e. the growth of the population, by an arrow to the right, i.e. towards
larger values of the population size N. Of course, this is not a complete description of our system, but it
gives a good idea about the behavior of our system. It shows that if we start at some initial value of N ∗ ,
then N will grow and the size of the population will be continuously increasing.

Definition 1 A phase portrait is the qualitative, graphical representation of the behavior of a system
described by a differential equation dx
dt = f (x). Note that such a phase portrait can only be drawn if f
does not depend on t.

Above, we obtained the phase portrait from the slope or rate of change of N(t). Now let us think
back a little. We obtained N(t) as the solution (integral) of dNdt = 4N. Thus, the slope, rate of change,
dN
or derivative of N(t) is simply dt = 4N! In other words, to construct the phase portrait of a general
differential equation dx
dt = f (x) one does not need to first obtain the solution x(t) and than find from this
solution its the slope. Rather, we can obtain this information directly from dx dt = f (x) and hence from
the function f (x). This means that we can apply the phase portrait analysis to any differential equation,
independent of whether it can be analytically solved.

To be able to construct a phase portrait of dx


dt = f (x), we need to know where to draw → or ← arrows on
the x-axis. The → arrow means growth of x, i.e. dx dx
dt > 0. The ← arrow means decline of x, i.e. dt < 0.
Because dxdt = f (x) we can simply draw the graph of function f (x) and then assign → to those regions of
x values for which the graph lies above the x-axis and ← to those regions for which the graph lies below
the x-axis.

Equilibria, stability, global plan

Let us consider two differential equations:

dx/dt = 4x (3.9)

and
dx/dt = 240 − 0.01x (3.10)
Let us sketch the phase portraits for (3.9) and (3.10).
38 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

x’ x’
x’<0 x’>0
f(x)
f(x)
x

a b
Figure 3.3:

For equation (3.9) dx/dt = 4x, the graph of f (x) = 4x is shown in the top left panel of Fig.3.3a and we
can draw → for x > 0 and ← for x < 0 (Fig.3.3a bottom). The interesting point here is x = 0, where f (x)
intersects the x-axis. For x = 0 we have dx/dt = 4x = 0. This means that we can not draw a → or ←
here. So what is the dynamics for x in this point? Since dx/dt = 0, the rate of change for x is zero, so x
does not change over time: if x = 0 x will always remain zero. Such points of a phase portrait are called
equilibria.

Definition 2 A point x∗ is called an equilibrium point of dx/dt = f (x), if f (x∗ ) = 0. In this point the
graph of f (x) intersects the x-axis.

We can also conclude from the phase portrait of eq.(3.9) in Fig.3.3 that if the initial value of x is to the
right (larger) or to the left (smaller) of the equilibrium point x = 0, the population size will go to plus or
minus infinity, respectively.

Now let us study equation (3.10). Again, our plan is: f (x) graph → phase portrait (fig.3.3b). Here we
have as an equilibrium point x = 24000 for which the function 240 − 0.01x intersects the x-axis. The
arrows (direction of change) for this case are shown in Fig.3.3b. So, here the dynamics will be as drawn
in Figure 3.4: for any initial condition, x will eventually approach the equilibrium point.
x
start b
equilibrium x
equilibrium
start a start b
start a t

Figure 3.4:

The equilibria of equations (3.9) and (3.10) are different. x diverges from the equilibrium of equation
(3.9) (left of the equilibrium arrows point leftward, right of the equilibrium arrows point rightward,
so away from the equilibrium). Such equilibrium points are called non-stable equilibria. In contrast,x
converges to the equilibrium point of equation (3.10) (left of the equilibrium arrows point rightward, right
of the equilibrium arrows point leftward, so towards the equilibrium). Such equilibrium points are called
stable equilibria or attractors (as they attract the value of the variable towards their particular value).
3.2 Q UALITATIVE ANALYSIS OF DIFFERENTIAL EQUATIONS 39

With this knowledge, let us take another look at Figure 3.3. We see that in Figure 3.3a, f (x) < 0 left of
the equilibrium, and f (x) > 0 right of the equilibrium. In other words the derivative of f , f 0 > 0 in the
equilibrium point. This produces an unstable equilibrium point. In contrast, in figure 3.3b, f (x) > 0 left
to the equilibrium and f (x) < 0 right to the equilibrium, and hence f 0 < 0 in the equilibrium point. This
produces a stable equilibrium point. Thus, we can derive the stability of an equilibrium point from f 0 ,
the slope of f in the equilibrium point.

Definition 3 An equilibrium point x∗ ( f (x∗ ) = 0) is stable if f 0 (x∗ ) < 0 and unstable if f 0 (x∗ ) > 0.

Now let us use our qualitative approach for a non-linear differential equation for which we can not find
the analytical solution: the logistic equation for population growth

dn/dt = rn(1 − n/K) n ≥ 0 (3.11)

This equation describes growth of a population in a medium with limited resources, i.e. the population
can not grow indefinitely, instead there is a maximum population size that can be supported. We can
study (3.11) for arbitrary values of parameters r, k. However for simplicity let us fix r = 0.5 and K = 10.
The equation becomes
dn/dt = 0.5n(1 − n/10) (3.12)
Let us find the phase portrait of (3.12). The graph of the equation 0.5n(1 − n/10) is a parabola open
in the downward direction (bergparabool), with roots (intersection points with the x-axis) at n = 0 and
n = 10 (see Fig. 3.5a). The phase portrait follows from this. The dynamics of the system is shown

f(n) n b
a

k n n c

A B n t

Figure 3.5:

in fig. 3.5b for different initial conditions. We see that for an initial population size 0 < n < 10 the
population increases to n = 10, whereas for an initial population size of n > 10 the population decreases
to n = 10. n = 10 thus is a stable equilibrium point and attractor of our system, whereas n = 0 is an
unstable equilibrium of the system.

We now learned that the slope of f (x), f 0 (x) in an equilibrium point x∗ determines the dynamics (di-
rection of arrows) around the equilibrium, and hence its stability. Another way of looking at this is as
follows. The function f (x) can be linearly approximated around the point x∗ as f (x) = f (x∗ ) + f 0 (x∗ ) ∗
(x − x∗ ). As x∗ is an equilibrium point ( f (x∗ ) = 0) this simply becomes f (x) = f 0 (x∗ ) ∗ (x − x∗ ). As
f 0 (x∗ ) is just anumber this is a linear differential equation, but not of x, but of the difference between
x and the equilibrium value x∗ ! Just as dx at
dt = ax has as a solution x(t) = Ae , we can now write for
d(x−x ) ∗ 0 ∗
dt = f 0 (x∗ ) ∗ (x − x∗ ) the solution (x − x∗ )(t) = Ae f (x )t . Now we see that if f 0 (x∗ ) < 0 we get a
declining exponential function, meaning that any difference between x and x∗ will disappear and the sys-
tem will return to the equilibrium x∗ . In contrast, if f 0 (x∗ ) > 0 we get an exponentially growing function,
40 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

meaning that differences between x and x∗ will grow and hence that the equilibrium is unstable. This
once more demonstrates the link between the dynamical behaviour described by the solutin and how this
same information follows from the phase portrait.

Differential equations can have multiple stable equilibria (attractors). For example, the model for the
spruce bud-worm population (3.13) has the following phase portrait (Fig.3.6).

du/dt = f (u) (3.13)

We see that there are two attractors: A1 and A2 which correspond to bud-worm populations of different

f(u)

u
u1 u2 u3

A1 A2 u

Figure 3.6:

size. We see that if the initial size of the population is 0 < u0 < u2 , then the population eventually reaches
A1; if u2 < u0 < ∞, then the population eventually reaches A2. These two intervals are called the basins
of attraction of the two stable equilibria, i.e. 0 < u0 < u2 is the basin of attraction for A1 and u2 < u0 < ∞
is the basin of attraction for A2. It is very important to know both the initial state and the equilibria and
basins of attraction of a system to be able to predict its future behavior.

Definition 4 The basin of attraction of a stable equilibrium point x∗ is the set of values of x such that, if
x is initially somewhere in that set, it will subsequently move to the equilibrium point x∗ .

Let us summarize the above into a general plan for finding a phase portrait:
1. Sketch the graph of f (x).
2. Determine where f (x) = 0 and draw equilibria points there.
3. Determine where f (x) > 0 and draw → there.
4. Determine where f (x) < 0 and draw ← there.
5. Determine attractors and their basin of attraction.

3.3 Systems with parameters. Bifurcations.

One of the aims of modeling in biology is to understand the behavior of a system for different conditions.
Note that in this case with different conditions we do not mean different initial conditions, i.e. different
initial population sizes, but we mean for example different environmental conditions in which a particular
species is living, or different types of species, which lead to for example different growth rates or carrying
capacities for the population we are interested in. In this case the differential equations will contain
parameters such as the growth rate k or the carrying capacity K which are not confined to a single specific
value but which can take on different values for the different populations. It then becomes important to
3.3 S YSTEMS WITH PARAMETERS . B IFURCATIONS . 41

find out whether the system behaves differently for these different possible values of k or K. For example,
it may be possible that for some values of k the population always goes to a stable equilibrium value,
whereas for other values the population always goes extinct.

If we work with differential equations containing parameters, for example dx/dt = rx(1 − x/K), always
be careful to distinguish the variables (x) from the parameters (r and K). We discussed this distinction
in Chapter 1. An example of an equation with parameters is the logistic equation for population growth
(3.11). This equation depends on two parameters r and K, where r accounts for the growth rate and
K accounts for the carrying capacity. Let us consider a slight modification of equation (3.11) for a
population which is subject to harvesting at a constant rate h:

dn/dt = rn(1 − n/k) − h (3.14)

where h is the harvesting rate (an extra parameter). Let us now fix the parameters r = 0.5 and k = 10
(the same values as in equation (3.12)), and study only the effect of varying the harvesting parameter h
on the dynamics of the population.

dn/dt = 0.5n(1 − n/10) − h (3.15)

When h = 0 equation (3.15) coincides with equation (3.12) which we studied in Fig.3.5. Now assume
that the harvesting h is not zero. Let us plot graphs of 0.5n(1 − n/10) − h for h = 0; h = 0.8; h = 1.6
(Fig.3.7a). The phase portraits for h = 0.0; h = 0.8; h = 1.6 are shown in Fig3.7b. We see that h > 0

f(n) h=0.0 f(n)

h=0.8
n n
h=1.6
h=0
h=0.8
a h=1.6 b c
Figure 3.7:

just shifts the parabola 0.5n(1 − n/10) downward. At h = 0.8 the behavior of the system is qualitatively
similar to the behavior of the system without harvesting (h = 0.0): the population eventually approaches
the stable non-trivial equilibrium. However the final size of the population in this case is smaller than
for the population without harvesting (fig.3.5a). At h = 1.6 the situation is different as we do not have
a stable and non-stable equilibrium anymore. The flow is now always directed to the left causing the
population to continuously decrease and eventually go extinct.

The biologically important question here is, what is the maximal possible harvesting rate at which the
population still survives. This critical harvesting is reached when the top of the parabola f (n) = 0.5n(1 −
n/10) exactly touches the n-axis (fig.3.7c), which occurs when the downward shift applied by h equals
the maximum height of the parabola. To find the maximum value of f (n), we first need to find the n
value for which it occurs: nmax. For this, remember that at the maximum of a function ( f (nmax)) its
derivative is zero ( f 0 (nmax) = 0). Thus, we find f 0 (n) and find for which n it equals zero. This value for
n we then substitute back into f (n) to give us the maximum level of f (n) and hence of h:

d(0.5n(1−n/10)) 2)
df
dn = dn = d(0.5n−5n
dn = 0.5 − 10n = 0 (3.16)
nmax = 5; f (nmax ) = 0.5 × 5 × (1 − 5/10) = 0.5 × 5 × 0.5 = 5/4
42 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

So the maximal value and therefore the maximal harvesting is h = 5/4

If h > 5/4 there are no equilibria and the population will go extinct. If h < 5/4 there is a stable and non-
stable equilibrium and the population will go to the stable equilibrium. At h = 5/4 we are at a boundary
between these two qualitatively different cases.

Definition 5 A bifurcation is a qualitative change in the system behavior, i.e. the number or the stability
of equilibria of the system changes

Bifurcations are studied in a special section of mathematics: theory of dynamical systems.

3.4 Extra: Solving linear differential equations

Let us consider again:


dN
= kN
dt
These types of equations can be solved using the method separation of variables. The main idea of this
method is to apply a somewhat dirty trick and think of dx
dt as a fraction of dx over dt. What the method
then does is move all N containing terms to one side and move all t containing terms to the other side of
the = sign. To do this for the above equation we divide left and right by kN and then multiply left and
right with dt. This results in the following equation:

dN
= dt
kN

We now integrate to t keeping in mind that the integral of 1/N is ln(N):

ln(N) = kt +C

To find N(t) now note that equation ln(x) = a is the same as x = ea :

N(t) = ekt+C = eC ekt = Aekt

with A = eC an arbitrary constant that depends on the initial size of the population.

Similarly we can also solve:


dN
= k − dN
dt
through the following sequence:
dN
= 1dt
Rk−dN
dN R
k−dN = 1dt
− d1 ln(k − dN) = t +C
ln(k − dN) = −dt − dC
(3.17)
k − dN = e−dt−dC = Ae−dt
−dN = −k + Ae−dt
dN = k − Ae−dt
N = dk − Ad e−dt
3.5 E XERCISES 43

If we now try to solve the general initial value problem N(0) we can rewrite this solution as follows:

N(0) = dk − Ad e−d0
N(0) = dk − Ad , or
(3.18)
A = k − dN(0), and
N = dk (1 − e−dt ) + N(0)e−dt

3.5 Exercises
0. Read the chapter
dx
(a) What does a differential equation dt = f (x) describe?

(b) What is the general solution of a differential equation?

(c) What is the solution of an initial value problem?

(d) What is a characteristic time?

(e) What is a phase portrait?

(f) What is an equilibrium?

(g) What is an attractor?

(h) What is a basin of attraction?

(i) What is a bifurcation?

1. Assume that a population grows according to the following equation:


dn
= 1.5n
dt
If the initial size of the population was n(0) = 30, find what will be the size of the population at
time t = 4. Find in how much time the population will double its initial size.
2. A bacterial population doubles its size every 20 minutes. The growth of this population N satisfies
the differential equation dN −1
dt = kN. Find the value of k in sec .
3. Proof that N(t) = dk (1 − e−dt ) + N(0)e−dt is the solution for the differential equation dN
dt = k − dN
for the case in which N(0) = 0
4. Study the listed differential equations by answering the following questions:

• Draw the phase portrait.

• How many equilibria do we have here? At which x?

• For each equilibrium tell whether is it stable or non-stable

• What will be the final value of x if t → ∞. (e.g. x converges to equilibrium, or x goes to


infinity, etc.)

• List attractor/attractors and determine their basin/basins of attraction.


dx
(a) dt = −15 + 8x − x2
dx
(b) dt = −4 + 5x − x2
44 D IFFERENTIAL EQUATIONS OF ONE VARIABLE

dx
(c) dt = −x(x2 + x − 6)
dx
(d) dt = 8x − x3
dx
(e) dt = f (x) with the following graphs of f (x):
f(x)

f(x)

0 1 6
x
0 4 10
x

dx 3x 2
(f) dt = 2+x 2 − x (this equation (Adler 1996) describes the dynamics of a population which
cannot bread successfully when numbers are either too small or too large)

5. The following equation describes the production of a gene product which concentration is denoted
by x:
dx/dt = −x(x − 0.2)(x − 1) + s
Here s is a parameter accounting for a chemical which produces the gene product. The initial state
of the system was x = 0, and the value of s = 0. Then, at some point in time the value of s was
slowly increased from s = 0 to s = smax and subsequently slowly decreased back to the value s = 0.

(a) Draw the graph of f (x) = −x(x − 0.2)(x − 1) + s for s = 0.

(b) Find the minimum value of the function f (x) = −x(x − 0.2)(x − 1) + s with s = 0.

(c) What will be the value of the concentration of the gene product x at the end of the described
process for smax = 0.005?

(d) The same for smax = 0.02?

(e) Show (graphically) that there is a critical value of smax that separates the qualitatively differ-
ent possible outcomes of this process. Find this critical value of smax .

6. Consider a model population with logistic growth which is subject to harvesting at a constant rate
h
dn/dt = rn(1 − n/k) − h (3.19)
Find the maximal yield h.
Chapter 4

Introduction to systems of two differential


equations

Thusfar we considered systems of a single differential equation, describing the dynamics of a single,
supposedly independent, variable. However, many biological systems need to be described by several
differential equations. As an example, to be able to describe the dynamics of a prey population we need
to take into account the dynamics of its predator and vice versa. The general form in which to write such
a system of two differential equations is:
 dx
dt = f (x, y) (4.1)
dy
dt = g(x, y)
here x(t) and y(t) are unknown functions of time t and f and g are functions of both x and y.

An example of a system of two differential equations is:


 dx
dt = ax + by (4.2)
dy
dt = cx + dy

the above equations only contain linear terms, so this is a linear system of differential equations.
Depending on the values and signs of the parameters a, b, c, d these equations can describe a range of
different processes. Let us consider the specific case a = −2, b = 1, c = 1, d = −2:
 dx
dt = −2x + y (4.3)
dy
dt = x − 2y

in this example x and y both decay with a rate −2, and are converted into one another with a rate 1.

Another example of a system of two differential equations is:


dx
= rx(1 − Kx ) − bxy (4.4)
dt
dy
= cxy − dy (4.5)
dt
As the terms rx2 , −bxy and cxy are non-linear, this is called a non-linear system. The system describes
the interplay between a prey population x and a predator population y. The prey population grows
logistically dx
dt = rx(1 − x/k), and is eaten at a rate −bxy. This means that the total number of prey killed
is proportional to both the number of predators and the number of prey, and hence that per predator the
46 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

number of prey killed is proportional to the number of prey. Later on in this course we will consider
more realistic relationships for this. The predator population grows proportional to the number of prey
eaten cxy and dies with a constant death rate d. Note that typically c < b: one eaten rabbit will not lead
to one new fox, and even one kg of eaten rabbit will not produce one kg of new fox. Let us consider the
specific case of r = 3, K = 1, b = 1.5, c = 0.5, d = 0.25:
 dx
dt = 3x(1 − x) − 1.5xy (4.6)
dy
dt = 0.5xy − 0.25y

4.1 Solutions of Linear 2D Systems

As for one dimensional systems, it is possible to find analytical solutions for linear two dimensional
systems (Eq. 4.2 and 4.3). Rather than deriving the solution for linear 2D systems here, we will simply
provide it and illustrate it’s analogy with the solution of linear one dimensional systems. Those students
interested in seeing how this solution is obtained are referred to the last section of this Chapter.

For one dimensional linear systems of the form:


dx
= ax (4.7)
dt
we learned that the general solution has the form:

x(t) = Aeat (4.8)

where A is an unknown constant depending on the initial value of x (A = x(0)).

In an analogous manner, a two dimensional system of the form:


dx
= ax + by (4.9)
dt
dy
= cx + dy (4.10)
dt
which in matrix notation can be written as:
 dx    
dt a b x
dy = (4.11)
dt
c d y

has as a general solution:

x(t) = C1 v1x eλ1t +C2 v2x eλ2t (4.12)


y(t) = C1 v1y eλ1t +C2 v2y eλ2t (4.13)

which in matrix notation can be written as:


     
x(t) v1x λ1 t v2x
= C1 e +C2 eλ2t (4.14)
y(t) v1y v2y

Note that similar to the single unknown A depending on x(0) in the one dimensional solution, we here
have two unknowns C1 and C2 depending on the initial values of x and y (x(0) = C1 v1x + C2V2x and
y(0) = C1 v1y+C2V2y ). Furthermore, λ1 , λ2 are the eigenvalues and v1 and v2 are the eigenvectors of the
a b
matrix A = . As we saw in Chapter 2, the eigenvectors indicate the major directions of change
c d
4.2 P HASE PLANE , T RAJECTORIES AND E QUILIBRIA 47

of the system described by the matrix. However, what turns out to be most important is the sign of the
corresponding eigenvalues, as these determine the type and stability of the equilibrium. How this works
we will discuss later on in this and the next Chapters.

Example
Let us find the general solution of system Eq. 4.3: The matrix A is here:
 
−2 1
A=
1 −2

Thus the characteristic equation is:

λ2 − (a + d)λ + (ad − bc) = λ2 + 4λ + 3 = 0

And we can write for the values of the eigenvalues:


p √
(−(a + d) ± (a + d)2 − 4(ad − bc)) (−4 ± 16 − 12)
λ1,2 = = = −2 ± 1
2 2
so λ1 = −1 and λ2 = −3.
Note that in this case we can also find the eigenvalues by rewriting the characteristic equation as:

(λ + 3)(λ + 1) = 0

For v1 we can now write:      


−b −1 1
v1 = = =
a − λ1 −1 1
and for v2 we can write:    
−b −1
v2 = =
a − λ2 1
So finally we can write the general solution as:
     
x(t) 1 −1t −1
= C1 e +C2 e−3t (4.15)
y(t) 1 1
or

x(t) = C1 e−1t +C2 e−3t (4.16)


y(t) = −C1 e−1t +C2 e−3t (4.17)

Note that it is not possible to find an analytical solution for the non-linear system 4.6, just as it was not
possible to find such solutions for non-linear one dimensional systems.

4.2 Phase plane, Trajectories and Equilibria

Having found the general solution for system Eq. 4.3 we can now use it to study the systems behavior.
Let us for example take as initial values x(0) = 1 and y(0) = 2 and plot x(t) and y(t) over the course of
time. How does this work? First, we determine from the initial values for x and y the values of C1 and C2
for this particular solution:

1= C1 e−1×0 +C2 e−3×0 (4.18)


2= −C1 e−1×0 +C2 e−3×0 (4.19)
48 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

2 2

x y

1 1

0 t 0 t
0 8 16 0 8 16

Figure 4.1:

as e0 = 1 we can simplify this into:


1= C1 +C2 (4.20)
2 = −C1 +C2 (4.21)
From the first equation we obtain C2 = 1 −C1 . Substituting this in the second equation gives us C1 = − 12
and hence C2 = 12 . Thus the particular solution can be written as:
x(t) = − 21 e−1t + 21 e−3t (4.22)
1 −1t 1 −3t
y(t) = 2e + 2e (4.23)
By filling in a series of values for t we now find the values of x and y at different points in time. Figure 4.1,
left shows the dynamics of x and Figure 4.1, right shows the dynamics of y as a function of time. Such a
time course followed by the system is called a trajectory. We see that both x and y decay exponentially
from their start value to the value 0.

2. 2.

y y

0 0

-2. -2.
-2. 0 2. -2. 0 2.
x x

Figure 4.2:

As the dynamics of x and y are co-dependent it would be nice to represent them in a single figure. One
solution would be to use a a 3D representation, with an axis for time, x and y. However, remember how
for one dimensional systems we depicted the dynamics using a single axis. On this axis, the location
itself indicates the x value and the direction of the local arrow indicates the direction of the dynamics
for that x value, thus getting rid of the time axis. Here we take a similar approach for two dimensional
systems. We use a two dimensional plane, and depict the dynamics of x along the x-axis, the dynamics
of y along the y axis, and use arrows to depict the direction of the dynamics over time. We call this plane
the phase plane or phase space.
4.2 P HASE PLANE , T RAJECTORIES AND E QUILIBRIA 49

Figure 4.2, left shows the same trajectory or system dynamics as shown in Figure 4.1 in this phase
plane representation. We again see how x and y decline from their initial values to the point (0, 0)
located in the center of the plane. Of course, having the general solution available, we can draw a
whole series of trajectories, starting from different initial conditions, to get a more general idea about the
systems dynamics. The result of this is shown in Figure 4.2, right. We see that independent of the initial
conditions, the trajectories go to the point (0, 0) suggesting that this is a stable equilibrium of the system.

1−
λ<0 λ>0
1−
t t
a b
Figure 4.3:

Note that we could have already predicted this overall behaviour from the solution of the system,
more specifically from its eigenvalues. For this, observe that the equations for x(t) and y(t) consist of
a sum of two exponential functions. The behavior of these exponential functions depends on the sign
of their exponent, if it is positive the exponential function grows to infinity, whereas if it is negative
the exponential function declines to zero (Figure 4.3). As here both eigenvalues are negative, both
exponential functions decline, causing x(t) and y(t) to decline to zero, independent from their starting
value.

Given that the point (0, 0) appears to be an equilibrium point of our system Eq. 4.3, let us now study more
in general the conditions for an equilibrium in two dimensional systems. As we saw for systems of one
equation, equilibria are those points in our system where the dynamics is stationary, i.e. nothing changes.
Therefore, for the 1D equation dx dx
dt = f (x) equilibria are determined as those points where dt = f (x) = 0,
i.e. where x does not change its value.

In 2D systems, the dynamics of x and y are co-dependent: the dynamics of x depends on both x itself
and on y, similarly, the dynamics of y depends on both y itself and x. This means that if we would find
dy
a point (x0 , y0 ) for which dx 0 0 0 0 0
dt = f (x , y ) = 0 but dt = g(x , y ) 6= 0, x would stay constant at x , while y
00 0 00
would change to y , causing the system to go to the point (x , y ). However, in this new point dx dt is no
0 0
longer equal to zero and x will also change. A similar story holds if we find a point (x , y ) for which
only dy 0 0 dx 0 0
dt = g(x , y ) = 0 and dt = f (x , y ) 6= 0. Thus, in order for the system to be stationary, that is for
dy
nothing to change both dx dt = 0 and dt = 0 should hold. Thus,

Definition 6 A point (x∗ , y∗ ) is called an equilibrium point of a system (4.1) if

f (x∗ , y∗ ) = 0, g(x∗ , y∗ ) = 0 (4.24)

Check yourself that for system 4.3 the point (0, 0) is the only equilibrium point.

As system 4.6 is a non-linear system it is not possible to find an analytical solution to determine its
dynamics. However, we can find the equilibria of system 4.6. For this we need to solve the following
system of algebraic equations: 
3x(1 − x) − 1.5xy = 0
(4.25)
0.5xy − 0.25y = 0
50 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

From the second equation we find y(0.5x − 0.25) = 0, which can be true either when y = 0 or when
x = 0.5. Substitution of y = 0 into the first equation yields 3x(1 − x) − 0 = 0. This equation has two
solutions x = 0 and x = 1. Substitution of the other case x = 0.5 into the first equation gives 3 ∗ 0.5 ∗ (1 −
0.5) − 1.5 ∗ 0.5y = 0, or y = 1. Thus we have found three equilibria points: (0, 0), (1, 0) and (0.5, 1). The
first equilibrium corresponds to a system with neither predators nor prey, the second to a system with
only prey, and the third to a system where predators and prey co-exist. The crucial question now is which
of these equilibria is stable, and hence to which situation will the system converge?

4.3 Nullclines and Vectorfields

As we saw in 1D systems, we can use a qualitative phase portrait analysis to determine the dynamics of
a system without needing its analytical solution. Let us develop a similar method for 2D systems.

For this let us first define so-called null-clines, lines in the phase plane at which only one of the two
variables of the system is in equilibrium:

Definition 7 The x-null-cline (or dx


dt = 0 null-cline) is the set of points satisfying the condition f (x, y) = 0.
dy
The y-null-cline (or dt = 0 null-cline) is the set of points satisfying the condition g(x, y) = 0.

Note that a single null-cline may consist of more than one line.

2 2

y y

0 0

-2 -2

-2 0 2 -2 0 2

x x

Figure 4.4:

Let us determine the null-clines for system 4.3. For the x null-cline we can write dx dt = −2x + y = 0,
dy 1
so y = 2x. For the y null-cline we can write dt = x − 2y, so y = 2 x. Figure 4.4, left shows the x and y
null-clines in the phase plane together with the trajectories that we found earlier. Based on this Figure
we can make a number of observations. First, we see that the equilibrium point (0, 0) corresponds to the
intersection point of the x and y null-clines. This makes perfect sense. Remember that for an equilibrium
dy
point both dx
dt = f (x, y) = 0 and dt = g(x, y) = 0 should hold. The first condition is the equation for the
x-nullclline, whereas the second condition is the equation for the y-nullcline. Thus, equilibrium points ly
on both the x and y null-clines and are hence intersection points of the nullclines.

Conclusion 1 Equilibria are the points of intersection of the x and y-null-clines.

Please note that this definition applies only to points of intersection of different null-clines.
4.3 N ULLCLINES AND V ECTORFIELDS 51

The second observation that we can make is that the null-clines separate the phase plane into distinct
regions in which trajectories move in qualitatively different directions. Figure 4.4, right shows the
different directions of the systems dynamics in a series of points located in the different regions of the
phase space. These direction vectors can be found as the direction of lines tangent to the trajectories in
Figure 4.4, left. This representation of the systems dynamics is called a vectorfield.

dx/dt>0 dx/dt<0
dy/dt>0 dy/dt>0
III IV
I II
dy/dt<0 dy/dt<0
dx/dt>0 dx/dt<0

Figure 4.5:

Why do null-clines divide the phase plane into regions with qualitatively distinct dynamics? Let us
assume that we are only interested in the qualitative dynamics of x and y, that is we are only interested
in whether they grow or decline but not in the exact rate at which they do so. When we study the system
in this qualitative manner there are only four possible types of dynamics (Figure 4.5):
I. x and y both increase
II. x decreases and y increases
III. x increases and y decreases
IV. x and y both decrease

I I
x’>0 y’>0
x’>0 y’>0 III
x’>0 y’<0

x’=f(x,y)=0
II
y’=g(x,y)=0

x’<0 y’>0

a b
Figure 4.6:

If we compare cases I and II we see that they differ by the sign of the dx dx
dt derivative: for case I dt > 0
dx dx
and for case II dt < 0. Therefore these two cases are separated by a boundary at which dt = 0 (fig.4.6a):
the x null-cline. Similarly, the transition from case I to case III occurs when dydt = 0 (fig.4.6b): the y
null-cline. Note that the x null cline not only defines the boundary between cases I and II, but also the
boundary between cases III and IV. Similarly, the y null-cline gives us not only the boundary between
caes I and III, but also the boundary between cases II and IV.

The third observation that we can make is that exactly at the null-clines one component of the vectorfield
is zero, causing the vectorfield to be either fully horizontal or vertical on the null-clines (Fig.4.6). Indeed,
on the x null-cline dx
dt = 0, so the horizontal component of the vector is absent, causing the vector to be
entirely vertical (Fig.4.6a). On a similar note, on the y null-cline dy dt = 0, so the vertical component is
absent, causing the vector to be entirely horizontal (Fig.4.6b).

The fourth and final observation we make is that in all 4 regions of the vectorfield the vectors point
52 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

toward the equilibrium point (0, 0), indicating that independent of where in the phase space the system
starts it will always converge to this stable equilibrium point. Thus, the phase plane with null-clines and
vectorfield give us the same information about the systems behaviour as the analytical solution (either
looking at the eigenvalues or using it to draw trajectories).

x0 (t)
 
Above we obtained the vectorfield from the solution trajectories, by determining the vectors
y0 (t)
 
x(t)
tangent to the solution trajectories . However, as x(t) is the integral of dxdt , the derivative of
y(t)
dy
x(t) is simply x0 (t) = dx
dt = f (x, y). Likewise, as y(t) is the integral of dt , the derivative of y(t) is simply
y0 (t) = dy dx
dt = g(x, y). Thus, we can directly find the vector in point a (x, y) from the values of dt = f (x, y)
and dydt = g(x, y) in that point ( f (xy) is the x or horizontal and g(x, h) gives the y or vertical component of
the vector). Note that while the computer often draws the two-dimensional resultant vector (see Figure
fig4.4), for us it is often more convenient to draw the horizontal and vertical components of the vector
separately (Figure 4.6).

x
0 0.5 1
Figure 4.7:

Now we can also draw a vectorfield for the non-linear system 4.6. For the dx dt = 0 null-clines we can write
3x(1 − x) − 1.5xy = x(3 − 3x − 1.5y) = 0. This equation has two solutions: x = 0 and y = 2 − 2x (the
two dashed lines in Fig.4.7). The dy dt = 0 null-clines we find from 0.5xy − 0.25y = y(0.5x − 0.25) = 0,
which also has two solutions: y = 0 and x = 0.5 (the solid lines in Fig.4.7). Now we find equilibria from
the intersection points of the x and y null-clines and obtain the previously found points (0, 0), (1, 0) and
( 21 , 1). Note that only intersection points of different null-clines are equilibria, so for example the point
(0, 2) is not an equilibrium.

Next we need to draw the vectorfield. The easiest way to do it is to first determine the vectorfield in one
region of the phase space and than use the null-clines to derive the vectorfield in the other regions. To
do this it is important to realize that at the x-null-cline the x-component of the vector field (horizontal)
changes its sign, and at the y-null-cline the y-component of the vector field (vertical) changes its sign.
To keep track of which null-cline corresponds to which variable and hence which vector field direction
(horizontal or vertical), it is practical to denote the x and y nullclines and the x and y components of the
vector field with different line types or colors, as we did in Fig.4.7.

So, let us first find the direction of the vectorfield in the point x = 1, y = 1. We find dx
dt = f (x, y) =
2 dy
3x − 3x − 1.5xy = 3 − 3 − 1.5 = −1.5, and dt = g(x, y) = 0.5xy − 0.25y. = 0.5 − 0.25 = 0.25. So x
4.3 N ULLCLINES AND V ECTORFIELDS 53

decreases and y increases and we need to draw the dashed arrow to the left and the solid arrow upward
(Fig.4.7). Next we complete the picture, finding the vectorfield in other regions and on the null-clines.
For example, to get into the region left from point (1, 1) we cross the solid line, so we need to change the
direction of the solid component of the vector. We finally get the picture of Fig.4.7.

Can we now tell from it the expected long term dynamics of the system, i.e. whether it goes to the
equilibrium (0, 0) where both predator and prey are extinct, or to the equilibrium (1, 0) in which only prey
are present, or to the equilibrium ( 21 , 1) where predator and prey co-exist? If we look at the vectorfield
close to (0, 0), we see that the vertical direction of the vectorfield points towards it, whereas the horizontal
direction points away from it. Apparently, the system moves closer to the equilibrium from the y direction
but moves further away from it in the x direction, suggesting that the equilibrium is unstable. For the
(1, 0) we have a similar situation, but with the approaching and diverging directions reversed. What
about equilibrium ( 21 , 1)? Unfortunately it is not easy to tell what happens here, the only thing we can
read from the vectorfield is a kind of rotating dynamics. So, although the null-clines and vectorfield
approach are often sufficient to determine the systems dynamics, this is not always the case! However, as
this system is non-linear we can also not find the general solution to determine the dynamics. In Chapter
6 we will discuss how we can still figure out the dynamics of the system in these cases.

General plan for phase portrait analysis

With this knowledge we can now write a general plan for finding the global phase portrait for 2D
systems. We assume that on the Oxy-plane the x-axis is the horizontal axis and the y-axis is the vertical
axis.

dy
1. Draw dx
dt = 0 null-clines from the equation f (x, y) = 0 using a dashed line and dt = 0 null-clines
from the equation g(x, y) = 0 using a solid line.
2. Choose a point in one of the regions in the x, y plane and find the x and the y -components of the
vector field. Use the dashed line for the x component and the solid line for the y component. For
finding the directions use the following rule: if f (x, y) > 0 the x component is directed as ’→’, if
f (x, y) < 0 it is directed as ’←’; if g(x, y) > 0 the y-component is directed as ↑, g(x, y) < 0 it is
directed as ↓.
3. Find the vector field in the adjacent regions using the following rule:

(a) change the direction of the dashed component of the vector field if in order to get to the
adjacent region you cross the dashed null-cline

(b) change the direction of the solid component of the vector field if in order to get tothe adjacent
region you cross the solid null-cline

(c) show the direction of the vector field on the null-clines.

4. Find the equilibria as intersection points of the dashed and solid null-clines
5. Try to find the dynamics around individual equilibrium points from the vectorfield
6. Try to derive from the dynamics around the individual equilibria the global dynamics of the system:
which equilibria are stable and what are their basins of attraction.
54 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

4.4 Systems with parameters. Bifurcations.

As we already discussed in Chapter 3, one of the major aims of using models in biology is to understand
whether a system will display qualitatively different behaviours under different conditions. In other
words, we want to know whether a change in parameter setting leads to a bifurcation of model behaviour.
Systems 4.3 and 4.6, that we discussed in detail in most of this Chapter were specific instances of the
systems 4.2 and 4.4, that is we filled in specific values for their parameters. Note that for drawing
trajectories, it is indeed necessary to choose specific parameter values. In contrast, for the method of
phase portrait analysis this is not necessary. However, one has to take into account that depending on the
(relative) values of parameters, different phase portraits may apply.

Let us illustrate this for a particular example. Let us assume we have the following system of two
differential equations: (
dx
dt = −ax + by
dy x2 (4.26)
dt = c x2 +d 2 − ey

Note that in this system, a, b, c, d and e all are parameters. First let us determine the null clines of this
system. For dx a
dt = 0 we obtain y = b x as the x null cline. This is a straight line going through the point
c x2
(0,0) and with a slope of ab . For dy dt = 0 we obtain y = e x2 +d 2 as the y nullcline. Recall that this is
a sigmoid function, going through point(0,0), reaching its half maximal y value at x = d and having a
maximal y value of ec .

Figure 4.8: Qualitatively different phase portraits for the system of Eq4.26.

In Figure 4.8 we have drawn two qualitatively different phase portraits. In Figure 4.8a the x and y
nullcline have a single intersection point at (0,0). In contrast, in Figure 4.8b the x and y nullcline have
a total of three intersection points. Which of these two situations occurs depends on the relative sizes of
the parameters. If we for example take constant values for parameters b, c, d and e, lowering the value of
parameter a will take us from the situation in Figure 4.8a to the situation in Figure 4.8b.

How to draw a vectorfield in this situation? As we do not know the values of the parameters, we can
dy
not simply fill in specific x and y values to find the values of dx dt and dt . The trick is to fill in either
“very small” or “very large” for x and y. As an example let us choose x very large and y very small.
This corresponds to the region under the sigmoidal y null-cline, as x can grow to infinity there while y is
limited by the null-cline: so x is large and y is small there. We now find that dx
dt = −a×big+b×small< 0,
cx2 dy
and that for x big the (x2 +d 2 ) part of dt approaches c, which will be bigger than ey for very small y, so
dx
dt > 0. After this we can find the rest of the vectorfield in the same manner as before.

From the vector field we observe that in the situation where equilibrium point (0, 0) is the only equilib-
rium, it is stable: vectors from all neighbouring regions of the vectorfield point towards it. In the situation
of Figure 4.8b we see that equilibrium point (0,0) is still stable, and that the third, highest equilibrium
point is also stable. In contrast, for the intermediate equilibrium point we see that the vectors in the
4.5 E XTRA : T HE GENERAL SOLUTION OF LINEAR 2D SYSTEMS 55

topleft and bottomright neighbouring regions of the vectorfield point towards it, whereas the vectors in
the topright and bottomleft regions point away from it. Put differently, this equilibrium point has a stable
direction through which the system approaches it and an unstable direction through which the system
diverges away from it. Note that the stable direction of this equilibrium creates a boundary, called a
separatrix, between the basins of attraction of the two other, stable equilibria (Figure 4.8c). This type
of equilibrium is called a saddle point and will be discussed in more detail in the next Chapter.

4.5 Extra: The general solution of linear 2D systems

We will not fully derive formula (4.14). Instead we will show the main ideas behind the real derivation.
For that let us first go back to the one dimensional case, and then extend it to a two dimensional system.
Thus, let us consider a simple 1D linear differential equation:
dx
= ax (4.27)
dt

In Chapter 3 we saw that we could solve (4.27) using the direct method of separation of variables and
subsequent integration. In contrast, here we will use another method, called the method of substitution.
The main idea is to substitute for the variable x a function for which we know that solutions of dx dt can
be described using that type of function. For 1D linear equations we learned in Chapter 3 that their
solutions can be described by exponential functions of the form Ceλt . Important questions such as: how
was this class of exponential functions found and is this class unique or would other functions also be
possible etc, will not be discussed here. Our aim here merely is to illustrate major components of the
solution approach. Once the class of functions is found, we need to check under which circumstances it
will satisfy the equations we are solving. Thus we will look for a solution of (4.27) of the form x = Ceλt ,
where C and λ are unknown coefficients. The main idea of the method of substitution is to find these
unknown coefficients for a particular system. Let us substitute x = Ceλt into (4.27). We find:

dx
= (Ceλt )0 = λCeλt
dt
or
λCeλt = aCeλt
We can cancel eλt and C, and we get:
λ=a (4.28)
Hence we found the following solution of (4.27):
x = Ceat (4.29)
where C is an arbitrary constant.

Now, let us use the same approach for the two dimensional system (4.2). It turns out that the class of
functions in two dimensions will be the same as in one dimension, but because we have two variables,
we need to introduce different constants for x and y, so our substitution will be
   
λt λt x Cx
x = Cx e ; y = Cy e or = eλt (4.30)
y Cy
where Cx ,Cy , λ are unknown coefficients. Let us make this substitution for a particular example:
 dx    
dt 1 4 x
dy = (4.31)
dt
1 1 y
56 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

Substitution for dx
dt = λCx eλt , dy λt λt λt
dt = λCy e , x = Cx e , y = Cy e gives:

λCx eλt = Cx eλt + 4Cy eλt
(4.32)
λCy eλt = Cx eλt +Cy eλt

we can cancel eλt , (but not Cx ,Cy as in one dimensional case), and get:

λCx = Cx + 4Cy
(4.33)
λCy = Cx +Cy

or in the matrix form:     


Cx 1 4 Cx
λ = (4.34)
Cy 1 1 Cy
We see that to find the solution of (4.31) we need to solve the problem (4.34), which is the eigen value
problem considered in Chapter 2. To solve it we find eigen values from the characteristic equation (2.22):
√ √

1−λ 4 2 − 2λ − 3 = 0 λ = 2± 22 +3∗4 = 1 ± 1 16 = 1 ± 2.
Det = (1 − λ) ∗ (1 − λ) − 4 = λ 12 2 2
1 1−λ
Hence λ1 = −1, λ2 = 3

We use the express formula (2.27) for eigen vectors and get:
         
v1x −4 −4 v2x −4
λ1 = −1; = = λ2 = 3; = . (4.35)
v1y 1 − (−1) 2 v2y −2

Formula (4.35) gives just one eigen vector for each eigen value. We also know (see formula (2.24 in
Chapter 2) that all eigen vectors can be found by multiplication of this eigen vector by an arbitrary
constant. If we denote by C1 the constant for the first eigen vector and by C2 the constant for the second
eigen vector we will get the following solution of the eigen value problem (4.34):
       
v1x −4 v2x −4
λ1 = −1; = C1 λ2 = 3; = C2 . (4.36)
v1y 2 v2y −2

If we substitute these eigen vectors into the formula (4.30) we find the following solutions of (4.31)
       
x −4 −1∗t x −4
= C1 e = C2 e3∗t (4.37)
y 2 y −2

Now let us prove the following property of system (4.2):


• If x1 , y1 and x2 , y2 are two solutions of (4.2), then x1 + x2 , y1 + y2 is also a solution of (4.2).
Proof: As x1 , y1 and x2 , y2 are the solution this means that they satisfy (4.2), i.e.
 dx1  dx2
dt = ax1 + by1 dt = ax2 + by2
dy1 dy2 (4.38)
dt = cx1 + dy1 dt = cx2 + dy2

If we add equations for dx dx2 dx1 dx2


dt and dt we get: dt + dt = ax1 + by1 + ax2 + by2 , which can be re-written
1

as: d(x1dt+x2 ) = a(x1 + x2 ) + b(y1 + y2 ). If we do the same for equations for dy 1 dy2
dt and dt we will finally get:
(
d(x1 +x2 )
dt = a(x1 + x2 ) + b(y1 + y2 )
d(y1 +y2 ) (4.39)
dt = c(x1 + x2 ) + d(y1 + y2 )

which explicitly shows that x1 + x2 , y1 + y2 is a solution of (4.2).


4.6 E XERCISES 57

If we apply this result for two found solutions (4.37) of (4.31) we can conclude that the sum of these two
solutions is also a solution of (4.31):
     
x −4 −4
= C1 e−1∗t +C2 e3∗t (4.40)
y 2 −2

We have proven formula (4.14) for a particular system. If we apply the same steps for a general system
(4.2) we will get the general solution in the form (4.14).

4.6 Exercises
0. Read the chapter

(a) What is the difference between a system of two differential equations and two separate
differential equations?

(b) When is a system of two differential equations linear?

(c) What do the C1 and C2 in the general solution stand for?

(d) What is a phase plane?

(e) What does a trajectory depict?

(f) When does a 2D system have an equilibrium?

(g) What is a null cline?

(h) What does a vectorfield depict? How does it relate to a trajectory?

(i) How are null clines and equilibria related?

(j) What is the difference between vectors in between and on top off null clines?

1. Find the solution for the following initial value problem:


 dx        
dt 1 −2 x x(0) 3
dy = =
dt
5 8 y y(0) −3

Hint: Proceed by first finding the general solution. After that, substitute t = 0 and x(0) = 3,
y(0) = −3 to find the values of constants C1 and C2 .
2. Two different concentrations of a solution are separated by a membrane through which the solute
can diffuse. The rate at which the solute diffuses is proportional to the difference in concentrations
between two solutions. The differential equations governing the process are:
(
dA/dt = − Vk1 (A − B)
dB/dt = Vk2 (A − B)

where A and B are the two concentrations, V1 and V2 are the volumes of the respective compart-
ments, and k is the rate constant at which the chemical exchanges between the two compartments.
If V1 = 20liters, V2 = 5liters, and k = 0.2 liters/min and if initially A = 3 moles/liter and B = 0,
find A and B as functions of time. Hint: this exercise is similar to the previous one!
58 I NTRODUCTION TO SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

3. Find equilibria of the following Lotka Volterra model with competition in the prey population.
Determine for which parameter values all equilibria are non-negative.

dN/dt = aN − eN 2 − bNP


dP/dt = cNP − dP a, b, c, d, e > 0

4. Protein synthesis depends on DNA transcription (a) making mRNA molecules (M) and translation
1
(c) of mRNA into proteins (P). Some proteins inhibit the transcription of their own mRNA ( 1+P ).
mRNA and proteins are degraded at rates b and d, respectively. This process gives the following
of two differential equations. Find equilibria of this model.
a

dM/dt = 1+P − bM P, M ≥ 0
dP/dt = cM − dP a, b, c, d > 0

5. Mathematical epidemiology also makes use of simple ODE models. One of these models describes
the number of susceptible individuals (S) and infected individuals (I). Individuals are born at rate
(B), and die at rate (µ). Susceptible individuals can become infected when they come into contact
with infected individuals (−γSI). Once infected, an individual has a certain death rate (α); this
may be different from the death rate of non-infected individuals. This process can therefore be
dS/dt = B − γSI − µS
modeled by the following equations:
dI/dt = γSI − αI B, α, γ, µ > 0
Find equilibria of this model. Determine for which parameter values all equilibria are non-
negative.
6. Draw the phase portrait of the system described in exercise 4. Include null-clines and vectorfield.
Determine (if possible) the type of equilibrium from the vectorfield.
7. Draw the phase portrait of the system described in exercise 5. Note that there are 2 possible
situations depending on whether both equilibria are non-negative or not. Determine if possible the
type of the equilibria in the two different situations.
Chapter 5

Equilibrium types in systems of two


differential equations

We saw that in one dimensional systems, there are only two types of equilibria, stable and unstable ones.
We can determine the equilibrium type from the phase potrait by seeing whether the arrows directly left
and right of the equilibrium point towards or away from the equilibrium. In two dimensional systems
there are a total of six different equilibrium types. As we saw in the previous Chapter, equilibria are
located at the intersection points of the x and y null-clines. These null clines divide the local phase space
around the equilibrium into four discinct regions. By looking at the directions of the vectorfield in these
four regions we can often determine the type and stability of the equilibrium. However, this is not always
possible.

In this Chapter we will illustrate all possible equilibrium types in linear systems, deriving both the ana-
lytical solution and the phase portrait. We illustrate when the phase portrait gives the same information
as the analytical solution and when the eigenvalues of the solution give more information. In the next
Chapter we will discuss how to find out about equilibria for which the phase portrait provides insufficient
information and the system is non-linear, so also no analytical solution can be found.

5.1 Real valued eigenvalues

Stable node

Let us start with the following system:


 dx
dt = −2x − y
dy (5.1)
dt = −x − 2y

It can be easily seen that this system has a single equilibrium point (0, 0). Note that this is always the
case for a system with only linear terms in x or y.
 
−2 −1
From the matrix A = , we can find the eigenvalues from the characteristic equation λ2 +
−1 −2
4λ + 5 = 0, which gives us (λ + 4)(λ + 1) = 0 and hence λ1 = −1 and λ2 = −4. So both eigenvalues
60 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

(a) (b)
2 2

y y

y
1 1

0 0

0 1 2 0 1 2
x x
x x

Figure 5.1:
       
−b 1 −b 1
are negative. Eigenvectors are v1 = = and v2 = = . From the
a − λ1 −1 a − λ2 2
shape of the general solution of a linear system:
     
x(t) v1x λ1 t v2x
= C1 e +C2 eλ2t (5.2)
y(t) v1y v2y
we learned that x(t) and y(t) depend on the sum of two exponential functions, in which the exponents
are the systems eigenvalues. As we saw in the previous Chapter, two negative eigenvalues result in two
declining exponential functions. Thus x and y decline to zero, causing (0, 0) to be an attractor of the
system. Indeed, in Figure 5.1a we see that independent of the initial conditions, all trajectories converge
to the point (0, 0)

Let us now find the same information using the method of phase portrait analysis. First we determine the
dy
null-clines of the system. From dxdt = −2x − y = 0 we find the x null-cline y = 2x. From dt = −x − 2y we
find the y null-cline y = 0.5x. These nullclines are drawn in Figure 5.1b. Next we find the vectorfield.
dy
In the point (1, 1) we find that dx
dt = −2 − 1 = −3 < 0 and dt = −1 − 2 = −3 < 0, so the vector points
to the left and downward. From this and the null-clines we find the rest of the vectorfield. We see that
in all four regions of the vectorfield the vectors point straight towards the equilibrium point, thus also
demonstrating that it is a stable equilibrium. So in this case vector field and analytical solution give the
same information.

Definition 8 A stable node is an equilibrium type for which the vectors in all four regions of the vec-
torfield around the equilibrium point towards it. As two diagonally opposing regions together constitute
one direction, along which either convergence towards or divergence from the equilibrium occurs, we
say that a stable node has two convergent or stable directions. In a linear system this corresponds to the
situation where the analytical solution has λ1 < 0 and λ2 < 0.

Unstable node

Let us now consider the alternative system:


 dx
dt = 2x + y
dy (5.3)
dt = x + 2y
So what is here the stability and type of the equilibrium point (0, 0)? For this system the eigenvalues
can be found from λ2 − 4λ + 3 = 0 or (λ − 3)(λ − 1) = 0, giving us λ1 = 1 and λ2 = 3. So now both
5.1 R EAL VALUED EIGENVALUES 61

(a) (b)
2 2

y y

y
1 1

0 0

0 1 2 0 1 2
x x
x x

Figure 5.2:

eigenvalues are larger than zero. This means that both exponential functions in the general solution (Eq.
5.2) grow to infinity, causing x and y to depart from the equilibrium point (0, 0) and grow to plus or minus
infinity. Indeed, in Figure 5.2a we see how all trajectories depart from the equilibrium point, indicating
it’s instability.

From 2x + y = 0 we find y = −2x for the x null-cline, and from x + 2y = 0 we find y = −0.5x for
the y null-cline, both similar to before. However, if we now compute the vector in the point (1, 1) we
dy
find dx
dt = 3 > 0 and dt = 3 > 0, completely the opposite as before. The null-clines and vectorfield are
shown in Figure 5.2b. We see that in all four regions of the vectorfield the vectors point away from
the equilibrium point, indicating that it is an unstable equilibrium. So, again, vector field and analytical
solution give the same information.

Definition 9 An unstable node is an equilibrium type for which the vectors in all four regions of the
vectorfield around the equlibrium point away from it. Put differently, it is an equilibrium type with two
unstable directions. In a linear system this corresponds to the situation where the analytical solution has
λ1 > 0 and λ2 > 0.

Saddle

Now we consider the system:  dx


dt = −x − y
dy (5.4)
dt = −2x − y
We find the eigenvalues from λ2 = 2λ − 1 = 0. In this case we can not easily √ find the eigenvalues
(−2± 4−4×−1) √
by factorizing this equation, and instead apply the abc-formula: λ1,2 = = −1 ± 2, so
√ √ 2
λ1 = −1 − 2 < 0 and λ2 = −1 + 2 > 0. So now the first eigenvalue is smaller and the second one is
larger than zero. This means that in the general solution (Eq. 5.2), the first exponential function declines
to zero, whereas the second exponential function grows to infinity. What effect will this have on the
dynamics of the system? The fact that one exponential function still grows to infinity suggests that in the
long run x and y will go the plus or minus infinity and hence go away from the equilibrium, just as was
the case for a system with two eigenvalues larger than zero. So what difference does having one negative
eigenvalue have? In Figure 5.3a we have drawn a range of trajectories for this system. We see that the
trajectories first converge to the equilibrium in one direction (corresponding the eigenvector belonging
to the negative eigenvalue), but subsequently diverge again from the equilibrium in another direction
(corresponding to the eigenvector belonging to the positive eigenvalue). Thus the systems behaviour is
62 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

(a) (b)
1 1

y y

y
0.5 0.5

0 0

0 0.5 1 0 0.5 1
x x
x x

Figure 5.3:

substantially different than for the first system with the two positive eigenvalues, but still the equilibrium
is unstable: the system moves away from it.

For the x null-cline we find y = −x and for the y null-cline we find y = −2x. In the point (1, 1) we find
dy
the vector dx
dt =< and dt =< 0. This allows us to determine the rest of the vector field. The null-clines
and vectorfield are shown in Figure 5.3b. We see that the vectors in the upper right and lower left regions
of the vectorfield point towards the equilibrium, whereas the vectors in the upper left and in the lower
right regions of the vectorfield point away from the equilibrium, thus giving very similar information as
the general solution and trajectories.

Definition 10 A saddle node is an equilibrium type for which the vectors in two oppsoing regions of
the vectorfield point toward the equilibrium and the vectors in the two other opposing regions of the
vectorfield point away from the equilibrium. Thus, the system has a stable and an unstable direction. In
a linear system this corresponds to the situation where the analytical solution has λ1 < 0 and λ2 > 0 or
λ1 > 0 and λ2 < 0.

5.2 Complex valued eigenvalues

The previous three equilibria types occurred in linear systems with real valued eigenvalues. Now we will
study what happens if complex valued eigenvalues appear.

Consider the general linear system:


 dx
dt = ax + by
dy (5.5)
dt = cx + dy
For the eigenvalues we can write the equation:
p
((a + d) ± (a + d)2 − 4(ad − bc))
λ1,2 = (5.6)
2

Let us now introduce D = (a + d)2 − 4(ad − bc) and rewrite this into:

((a + d) ± D)
λ1,2 = (5.7)
2
5.2 C OMPLEX VALUED EIGENVALUES 63

What happens if D < 0? In Chapter 2 we learned that complex numbers were invented to still√allow
us
√ to solve the√equation.
√ √ we can rewrite D (remember D < 0) as −1 × −D and hence D as
Indeed
−1 × −D = −1 −D = i −D and hence:

((a + d) ± i −D)
λ1,2 = (5.8)
2

(a+d) −D
If we now call 2 = α and 2 = β this gives us:

λ1,2 = α ± iβ (5.9)
Thus, the two eigenvalues have the same real part α and an opposite imaginary part iβ, and are hence
each others complex conjugates.

Because eigenvalues are complex so are their corresponding eigenvectors. For example:
       
−b −b −b 0
λ1 = = = +i (5.10)
a − λ1 a − (α + iβ) a−α −β

so we can write v1 = vr1 + ivi1 , where v1r is the real part and iv1i is the imaginary part of the vector.

If we put the complex eigenvectors and values in the equation for the general solution we get:
 
x(t)
= C1(v1r + iv1i )e(α+iβ)t +C2(v2r + iv2i )e(α−iβ)t (5.11)
y(t)

but it is still quite unclear what this means for the behavior of the solutions.

To get an idea about this we need to introduce a new mathematical relationship. Just as we have equations
relating trigonometric functions (e.g sin2 x + cos2 x = 1) or exponential functions (e.g. ea+b = ea × eb ),
there is also a special function relating trigonometric and exponential functions via complex numbers:

eix = cosx + isinx (5.12)

(or e−ix = cosx − isinx), this famous equation is called Euler’s formula. With this formula we can rewrite:

eα±iβ = eα e±iβ = eα (cosβ ± isinβ) (5.13)

and hence:
 
x(t)
= C1(v1r + iv1i )eαt (cosβt + isinβt) +C2(v2r + iv2i )eαt (cosβt − isinβt) (5.14)
y(t)

Which can be further rewritten into:


 
x(t)
= eαt {C1(v1r + iv1i )(cosβt + isinβt) +C2(v2r + iv2i )(cosβt − isinβt)} (5.15)
y(t)

Note that at this point the general solution would give us both real valued and complex valued x and y
values. This means that some further mathematical operations are needed to substract the subset of real
valued solutions from the larger set of total solutions. However, we can already get a clear idea about the
dynamics of the system from the total set of solutions described by the above equation. Remember that
for real valued eigenvalues we have a solution in which x and y depend on the sum of two exponential
functions, whose exponents are the two eigenvalues. In contrast, for complex valued eigenvalues we see
64 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

(a) (b) (c)


2 3 2

y y

x, y
y

y
0 0 yx 0

-2 -3 -2

-2 0 2 0 2.5 5 7.5 10 -2 0 2
x x
t t
x x

Figure 5.4:

here that x and y are the product of a single exponential function with a complex sum of sine and cosine
functions. These sine and cosine functions arise from the complex part of the eigenvalues (i.e. they
contain the β) and cause oscillatory dynamics of x and y over time. The exponential function arises from
the real part of the eigenvalues (i.e it has α as an exponent) and scales the amplitude of these oscillations.
Thus, it will be the sign of the real part α that determines whether oscillations blow up and the equilibrium
is unstable, or that oscillations damp out and the system converges to a stable equilibrium.

Let us now discuss the types of equilibria that can occur under these conditions for specific systems.

Stable spiral

Consider system:
 dx
dt = −x + 2y
dy (5.16)
dt = −2x − y

From the system we obtain the characteristic equation λ2 + 2λ + 5 = 0, which gives us λ1,2 = −1 ± i2.
Thus, eigenvalues are complex, with a negative valued real part. Due to the complexity we expect
oscillatory dynamics, and due to the negative real part we expect damping of the oscillations producing
eventual convergence to the equilibrium (0, 0). Figure 5.4a shows trajectories of the system. We indeed
see an inward spiralling motion towards the equilibrium. Figure 5.4b shows the dynamics of x and y as a
function of time rather than in the xy plane, now we indeed see an oscillatory dynamics with decreasing
amplitude.

Next we study whether we can obtain the same information from a phase portrait analysis. We find
for the x null-cline y = 21 x and for the y null-cline y = 2x. For the point (1, 1) we find dx
dt = 1 > 0 and
dy
dt = −3 < 0, allowing us to draw the vectorfield. In Figure 5.4c the resulting phase portrait is shown.
From this we see that the dynamics undergo a rotating motion. However, we can not tell from the phase
portrait whether this rotating motion is converging to the equilibrium, diverging from it, or staying at
a constant distance. Thus in this case the eigenvalues give us more information than the phase portrait
analysis.
5.2 C OMPLEX VALUED EIGENVALUES 65

(a) (b) (c)


2 500 2

y y x

x, y
y

y
0 0 0

-2 -500 -2

-2 0 2 0 3.25 6.5 9.75 13 -2 0 2


x x
t t
x x

Figure 5.5:

Definition 11 A stable spiral is an equilibrium type for which the trajectories spiral inward towards
the equilibrium (either clockwise or anticlockwise, depending on the vectorfield). In a linear system it
corresponds to the situation where the analytical solution has complex eigenvalues λ1,2 = α ± iβ, and
α < 0.

Unstable spiral

Next we consider the system:


 dx
dt = x + 2y
dy (5.17)
dt = −2x + y

where we obtain from the characteristic equation λ2 − 2λ + 5 = 0 the eigenvalues lambda1,2 = 1 ±


i2.Again, eigenvalues are complex, but now with a positive real part. Indeed now we see the trajectories
spiralling out from the equilibrium, as one would predict for a positive real part of the complex eigenval-
ues (Figure 5.5a). Figure 5.5b shows the dynamics of x and y over time, showing oscillatory dynamics
with a very rapidly increasing amplitude.

We obtain for the x null-cline y = − 21 x and for the y null-cline y = 2x. The vectorfield in point (1, 1)
dy
is dx
dt = 3 > 0 and dt = −1 < 0, from which we obtain the rest of the vectorfield. Figure 5.5c shows
the null-clines and vector field. Note the similarity with the vectorfield in Figure 5.4c. However, the
trajectories in Figure 5.5a look very different from those in Figure 5.4a. Indeed, from the phase portrait
we can not distinghuish whether the rotating dynamics will spiral inward or outward, we need to know
the eigenvalues to determine this.

Definition 12 An unstable spiral is an equilibrium type for which the trajectories spiral away from
the equilibrium (either clockwise or anticlockwise, depends on the vectorfield). In a linear system it
corresponds to the situation where the analytical solution has complex eigenvalues λ1,2 = α ± iβ, and
α > 0.
66 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

(a) (b) (c)


2 2 2

y y

x, y
y

y
0 0 0

y
x

-2 -2 -2

-2 0 2 0 5 10 15 20 -2 0 2
x x
t t
x x

Figure 5.6:

Center point

Finally consider the system:


 dx
dt = x + 2y
dy (5.18)
dt = −2x − y


From which we obtain the characteristic equation λ2 + 3 = 0 giving us eigenvalues λ1,2 = ±i 3. So now
we have complex eigenvalues of which the real part is zero. This means that the exponential function
is equal to one, and hence is neither blowing up (α > 0) nor squeezing off (α < 0) the sine and cosine
functions. Thus, we should expect neither divergence from nor convergen to the equilibrium. Figure
5.6a shows a series of trajectories for this system. We see indeed neither convergence nor divergence,
instead the starting point of the trajectory defines the starting distance to the equilibrium and this distance
is maintained by the trajectory. As different trajectories start at different distances from the equilibrium
point the system ends up cycling at different distances from the equilibrium. Figure 5.6b shows the time
dynamics of x and y for one particular trajectory. We indeed see a constant amplitude oscillation for both
variables.

The system has an x null-cline y = − 12 x and an y null-cline y = 2x. In the point (1, 1) the vectorfield is
dx dy
dt = 3 > 0 and dt = −1 < 0, from which we can obtain the rest of the vectorfield. As before, from the
thus obtained phase portrait (Figure 5.6c) we can only determine a rotating dynamics, but not whether
it converges, diverges or maintains a certain distance to the equilibrium. Thus, for all these three cases
eigenvalues are more informative than a phase portrait.

Definition 13 A center point is an equilibrium type for which the trajectories rotate at a constant dis-
tance from the equilibrium (either clockwise or anticlockwise, depends on the vectorfield). This distance
is determined by the initial condition of the system. In a linear system it corresponds to the situation
where the analytical solution has complex eigenvalues λ1,2 = α ± iβ, and α = 0.
5.3 I NSUFFICIENT INFORMATION IN THE PHASE PORTRAIT 67

(a) (b)
2 2

y y

y
1 1

0 0

0 0.5 1 0 0.5 1

(c) x (d) x
2 2

y y
y

y
1 1

0 0

0 0.5 1 0 0.5 1

(e) x (f) x
2 2

y y
y

1 1

0 0

0 0.5 1 0 0.5 1
x x
x x

Figure 5.7: Qualitatively different steady states determined by the local vector field in the four regions
defined by the nullclines. Spiral points (a-d): the vector field suggests rotation. The spiral point in (a,b)
is stable which can be guessed because increasing x at the steady states makes dx/dt < 0 and increasing
y at the steady states makes dy/dt < 0 (which is stabilizing). The spiral in (c,d) is unstable which can be
guessed because increasing y at the steady states makes dy/dt > 0 (which is destabilizing). Panels (e &
f) illustrate that nodes can also have a rotating vector field, i.e., that one cannot tell with certainty from
the local field whether or not a steady state is a spiral point.

5.3 Insufficient information in the phase portrait

As we already saw in the previous section, the phase portrait does not always contain sufficient infor-
mation to determine the equilibrium type and long term dynamics of the system. This is once more
illustrated in Figure 5.7. We see that a rotating dynamics in the vectorfield may occur for stable and
unstable spiral equilibria (first two rows), but also for center point equilibria (Figure 5.6), and even for
stable or unstable node equilibria (bottom row) in certain cases.

However, one can sometimes still get more information about the stability of an equilibrium from the
local vector field. In 5.7c, for instance, and one can see that increasing y from its steady state value
68 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

makes dy/dt > 0. Locally, there must be some positive feedback allowing y to increase further when
y increases. This is definitely destabilizing, and the trajectory in 5.7d confirms that this is an “unstable
spiral” point.

Conversely, the spiral point in 5.7a is stable, and locally has negative feedback for both x and y (i.e.,
increasing x makes dx/dt < 0 and increasing y makes dy/dt < 0), which has a stabilizing influence.
Because the stability of spiral points also depends on the local difference in time scales of the horizontal
(x) and vertical (y) variables, the local vector field is not always sufficient to determine the stability of
the steady state. Even the suggestion of rotation in a local vector field is not sufficient to determine with
certainty that the steady state is a spiral. 5.7e & f show that the same vector field defining a stable spiral
point in 5.7a can actually also correspond to a stable node.

Still, in some cases vector field information alone will be insufficient to determine the equilibrium type
or stability. In those cases further analysis will be needed, which will be discussed in the next Chapter.

5.4 Equilibrium types and stability

We will call an equilibrium point stable, if there is a neighborhood of this equilibrium, such that all
trajectories which start in this neighborhood will converge to the equilibrium (Fig.5.8a). We will call the
equilibrium point non-stable, if there is at least one diverging trajectory from the close neighborhood of
this equilibrium (fig.5.8b).

y y

x x
a b
Figure 5.8:

If we analyze the stability of the 6 types of equilibria studied in the previous section we find the following:

Stable equilibria

1. Stable node λ1 < 0; λ2 < 0 real


2. Stable spiral λ1,2 = α ± iβ; α < 0

Non-stable equilibria

1. Non-stable node λ1 > 0; λ2 > 0 real


5.5 E XERCISES 69

2. Non-stable spiral λ1,2 = α ± iβ; α > 0


3. Saddle point λ1 < 0; λ2 > 0; or λ1 > 0; λ2 < 0 real

In case of spirals and nodes the stability and non-stability is obvious. In case of a saddle point we have a
converging trajectory, however the existence of the diverging trajectory implies that this equilibrium point
is non-stable. The last case,λ1,2 = ±iβ (center point), is non conclusive. Trajectories do not converge
and do not diverge from the equilibrium. Usually this case is treated as neutrally stable. All these cases
can be summarized in the following theorem.

Theorem 1 If all eigenvalues of the linear system (6.13) have negative real parts, then the equilibrium
point x = 0, y = 0 is stable.

It is easy to see, that this theorem includes all listed stable equilibria. It obviously works for a stable
spiral, but it also works for a stable node, because any real number can be considered as a complex
number with imaginary part equal to zero. For example: −3 can be represented as z = −3 = −3 + i0,
and Rez = −3; Imz = 0. Thus, in contrast to the graphical, qualitative null-cline and vector field method,
using this analytical method we can always determine the type and stability of the equilibrium.

5.5 Exercises
0. Read the chapter
(a) Which equilibrium types are there in 2D systems?

(b) Which of them are stable?

(c) How can you tell the equilibrium type from the general solution?

(d) When do we get rotating dynamics?

(e) When is the vectorfield not enough to determine equilibrium type?


1. Find eigen values and eigen vectors (vectors only for real eigen values) of the following systems.
Determine equilibrium type and sketch phase portraits.
 dx
(a) dt = x + 4y
dy
dt = 2x + 3y
 dx
(b) dt = 5x − y
dy
dt = 3x + y
 dx
(c) dt = 3x − 5y
dy
dt = x − y
 dx
(d) dt = −2x + y
dy
dt = x − 2y
 dx
(e) dt = −2y
dy
dt = x − 2y
 dx
(f) dt = −x − y
dy
dt = 2x + y
70 E QUILIBRIUM TYPES IN SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

 dx
dt = −2x − y
(g) dy
dt = 3x + 2y

2. Draw the phase portrait of the following systems of differential equations. Explain their dynamics.
 dx
(a) dt = x(1 − x − y) x ≥ 0; y ≥ 0
dy
dt = y(1 − 2x)
 dN 2
dt = 2N − NP − N
(b) dP 2 N ≥ 0; P ≥ 0
dt = 3P − 2NP − P

3. Study the following linear system with a parameter a:


 dx
dt = −2x − ay
dy
dt = 3x − y

Find the types of equilibrium which are possible for different values of −∞ < a < ∞. Give the
parameter region for each equilibrium and draw qualitative phase portraits. For which parameter
values is the equilibrium stable?
4. For which values of parameters a and b does the following system have periodic oscillations due
to a center point equilibrium:  dx
dt = −ax + y
dy
dt = (2a − 3)x − by
5. Compartmental models play an important role in different parts of population biology, pharma-
cology and biochemistry. They describe the interaction between several processes, which may be
interactions of populations, chemical reactions, etc. A two compartment model is schematically
shown in fig.5.9. It represents two interacting species x and y. The concentration of the species
x can be changed either due to a transition to species y with the rate given by ax, or x can decay
with the rate cx. Similar transitions exist also for the species y. The rates of these processes are
specified in the figure.

ax
x y
by
cx ey

Figure 5.9:

(a) Derive a system of two differential equations for species x and y.

(b) If a = 0.5, b = 2, c = 4.5, e = 3 find equilibrium type. Is it stable?

(c) Determine the stability of the equilibrium in a general case when a > 0, b > 0, c > 0, e > 0
Chapter 6

Linear approximation of non-linear 2D


systems

In the previous chapter we saw that the dynamics and equilibrium types of a 2D system can often be
determined using a phase portrait analysis, but that this approach does not always work. We saw that
in cases where the phase portrait was insufficient the eigenvalues of the analytical solution allow us
to determine the systems dynamics. However, for non-linear systems we can not find the analytical
solution. In this Chapter we will discuss how to analyse the dynamics of a system for which both the
phase portrait aproach and finding an analytical solution fails. We show that, using the the technique of
linear approximation, we can find a linear system that approximates our original system locally arount
the equilibrium point, and which analytical solution approaches the behavior of the original system close
to the equilibrium point.

6.1 Partial derivatives

The function f (x) = ax2 + b can be represented in a two dimensional plot, with the value of x on the
x-axis and the value of f (x) on the y-axis. By taking the derivative of f in a point x∗ , f 0 (x∗ ), we obtain
the slope of the (line tangent to the )graph of f (x) in point x∗ . Because the function f (x) only depends
on x we call f 0 (x) the derivative of f (x).

Figure 6.1:

Now let us assume we have a function f (x, y) = 3x − 3x2 − 1.5xy. To plot it we need a threedimensional
72 L INEAR APPROXIMATION OF NON - LINEAR 2D SYSTEMS

graph, with x on the x axis, y on the y axis and the value of the function f (x, y) on the z axis (Figure
6.1). In this case the function depends on both x and y, and therefore we speak of partial derivatives.
Just as the derivative f 0 (x) = d f /dx of f (x) describes how f changes as x changes the partial derivative
∂ f /∂x describes how f changes as x changes and y does not change, whereas the partial derivative ∂ f /∂y
describes how f changes as y changes and x does not change.

The main idea of finding a partial derivative of a function f (x, y) is to fix one variable at a constant value,
say x = x∗ . After that we will have a function with only one variable y left ( f (x∗ , y)). Now, we can find
the derivative of f (x∗ , y) (with respect to y as x is constant) as the usual derivative of a function of one
variable y. For example, f (x, y) = 3x − 3x2 − 1.5xy. Let us fix x = x∗ = 2. We get the following function
of one variable: f (2, y) = 3 ∗ 2 − 3 ∗ 22 − 1.5 ∗ 2y = −6 − 3y. We can easily find the derivative now:
d f (2, y)/dy = d(−6 − 3y)/dy = −3. We can denote it as ∂ f /∂y|x=2 = −3. For the partial derivative of
f with respect to y for any constant value of x we can write: ∂ f /∂y = ∂(3x − 3x2 − 1.5xy)/∂y = −1.5x.

Example. Find ∂z/∂x and ∂z/∂y for z = y3 sin x

Solution ∂z/∂x = y3 cos x, as for ∂/∂x we fix y, and ∂(sin x)/∂x = cos x. Similarly, ∂z/∂y =
∂(y3 sin x)/∂y = 3y2 sin x, as x and hence sin x is treated as a constant.

y x=x*
y* y=y*

x
x*

Figure 6.2:

The geometrical representation of a partial derivative is clear from fig.6.2. To compute ∂ f /∂x we fix
y, i.e. assume that y has some value y = y∗ . The condition y = y∗ geometrically gives a horizontal line
on the Oxy plane fig.6.2a, or a line parallel to the x-axis. In 3D this line gives a curve on the 3D surface
in graph fig.6.2b, which is a 1D function. The partial derivative with respect to x for this particular y∗
will give us the slope of the tangent line to this 1D function. Thus (see fig.6.2) ∂ f /∂x gives the slope of
the tangent line in the direction of the x-axis or the rate of change of f (x, y) in the x direction. Similarly,
computing ∂ f /∂y we fix x, i.e. assume that x has some value x = x∗ . It gives us a vertical line on the
Oxy plane fig.6.2a, or a line parallel to the y-axis. Thus ∂ f /∂y gives the slope of the tangent line in the
direction of the y-axis, or the rate of change of f (x, y) in the y direction. If we consider f (x, y) as a
mountain ∂ f /∂x gives the slope of the mountain if we climb in the x-direction and ∂ f /∂y gives the slope
of the mountain if we climb in the y-direction.

Note, that in general at each point on a surface we can draw a tangent line in any direction. The partial
derivatives ∂ f /∂x and ∂ f /∂y give us the slopes for the extreme cases, when we either only move in the
x or only move in the y direction. The slope of a tangent line in any other direction can be obtained as a
weighted combination of these two slopes.
6.2 L INEAR APPROXIMATION 73

6.2 Linear approximation

df f (x)− f (x∗ )
In Chapter 1 we saw that the derivative of f (x) at point x∗ can be written as f 0 (x∗ ) = dx = lim∗ x−x∗ .
x→x
We subsequently showed that this expression can be used to write a linear approximation of f (x) for x
close to x∗ as f (x) ∼ f 0 (x∗ )(x − x∗ ) + f (x∗ ). Let us now try to derive a similar function for the linear
approximation of a function f (x, y). Let us assume that we know f (x, y) and its partial derivatives ∂ f /∂x
and ∂ f /∂y at some point x∗ , y∗ and that we want to approximate the value of the function f at a nearby
point x, y (fig.6.3).

y
(x,y)
∆ f2
(x*,y*)
(x,y*)
∆ f1
x
Figure 6.3:

Let us move to the point x, y in two steps. Let us first move from the point x∗ , y∗ to the point x, y∗ ,
i.e. in the x-direction, and then from x, y∗ to x, y, i.e. in the y-direction. Let us apply the formula for
approximation of a function of one variable in formulation (1.2) at both parts of this motion. In the first
step we move in the x direction so the change in function value (∆ f1 ) will be the rate of change of the
function in the x direction (∂ f /∂x) times the distance travelled between the points (x − x∗ ):

∆ f1 = f (x, y∗ ) − f (x∗ , y∗ ) = (∂ f /∂x)(x − x∗ ) (6.1)

Similarly, in the second step, we move in the y-axis direction. Therefore the change of the function value
(∆ f2 ) now is the the rate of change of the function in the y direction (∂ f /∂y) times the distance travelled
between the points (y − y∗ ):

∆ f2 = f (x, y) − f (x, y∗ ) = (∂ f /∂y)(y − y∗ ) (6.2)

If we add equations (6.1) and (6.2) and solve it for f (x, y) we find the following formula which gives the
approximation for a function of two variables:

f (x, y) ≈ f (x∗ , y∗ ) + (∂ f /∂x)(x − x∗ ) + (∂ f /∂y)(y − y∗ ) (6.3)

This expression is called a linear approximation, as the independent variables x, y are in the first power
only, we do not have terms x2 , y2 , or xy, or etc.

Example Find the linear approximation for the function ex+2y at the point x = 0, y = 0

Solution. We use the formula (6.3) with f (x, y) = ex+2y and x = 0, y = 0.

f (x, y) = e(0+0) = 1;

∂ f /∂x = ex+2y ; at x = 0, y = 0, ∂ f /∂x = e0+2∗0 = 1


74 L INEAR APPROXIMATION OF NON - LINEAR 2D SYSTEMS

∂ f /∂y = ∂(ex+2y )/∂y = ex+2y ∗ ∂(x + 2y)/∂y = 2ex+2y , at x = 0, y = 0, ∂ f /∂y = 2e0+2∗0 = 2

Finally, ex+2y ≈ 1 + 1 ∗ x + 2 ∗ y.

At x = 0.1, y = 0.1 the approximate formula gives: ex+2y ≈ 1 + 1 ∗ 0.1 + 2 ∗ 0.1 = 1.3. The exact value
of ex+2y = e0.3 = 1.3498

6.3 Linearization of a system: Jacobian

Consider a general system of two differential equations:


 dx
dt = f (x, y) (6.4)
dy
dt = g(x, y)

Assume that system (6.4) has an equilibrium point at (x∗ , y∗ ).

f (x∗ , y∗ ) = 0

(6.5)
g(x∗ , y∗ ) = 0

Let us first find a linear approximation of f (x, y) close to the equilibrium:

f (x, y) ≈ f (x∗ , y∗ ) + (∂ f /∂x)(x − x∗ ) + (∂ f /∂y)(y − y∗ )

As (x∗ , y∗ ) is an equilibrium, we know that f (x∗ , y∗ ) = 0 so we get

f (x, y) ≈ (∂ f /∂x)(x − x∗ ) + (∂ f /∂y)(y − y∗ ) (6.6)

A similar approach for g(x, y) yields:

g(x, y) ≈ (∂g/∂x)(x − x∗ ) + (∂g/∂y)(y − y∗ ) (6.7)

If we now replace the right hand sides of (6.4) by their approximations (6.6), (6.7), we get:
∗ ∗
 dx
dt = (∂ f /∂x)(x − x ) + (∂ f /∂y)(y − y ) (6.8)
dy ∗ ∗
dt = (∂g/∂x)(x − x ) + (∂g/∂y)(y − y )

The system (6.8) is simpler than the original system (6.4), as the partial derivatives in (6.8) are con-
stants/numbers (as they are evaluated at the equilibrium point x∗ , y∗ ). So we can rewrite (6.8) as :
∗ ∗
 dx
dt = a(x − x ) + b(y − y ) (6.9)
dy ∗ ∗
dt = c(x − x ) + d(y − y )

where a = ∂ f /∂x; b = ∂ f /∂y; c = ∂g/∂x, d = ∂g/∂y. As x∗ and y∗ are constants, and hence their deriva-
dx∗ d(x−x∗ ) d(y−y∗ )
tives are zero, we can also write dx dx
dt = dt − dt = dt and similarly dy
dt = dt giving us
d(x−x∗ )
(
∗ ∗
dt ∗ = a(x − x ) + b(y − y ) (6.10)
d(y−y )
dt = c(x − x∗ ) + d(y − y∗ )

Note that this system is exactly the same as a general linear system, except that here it says x − x∗ and
y − y∗ everywhere where in a usual linear system it says x and y. Therefore, we can write the solution of
system 6.10 as:
(x − x∗ )(t)
     
v1x λ1 t v2x
= C1 e +C2 eλ2t (6.11)
(y − y∗ )(t) v1y v2y
6.3 L INEARIZATION OF A SYSTEM : JACOBIAN 75

What does this mean? Recall that in a linear system there is always a single equilibrium at the point
(0, 0). Therefore the x and y value in a point is automatically also the distance in the x and y direction
to the equilibrium point (0, 0). From the sign of the eigenvalues in the exponential functions of the
solutions we can determine whether the x and y values grow away from or decline towards zero, and
hence whether the distance to the equilibrium grows or declines. Now, our system 6.10 is a linearization
of a non-linear system with an equilibrium at the point (x∗ , y∗ ). The solution 6.11 now does not state how
the x and y values evolve over time, but rather states how the distance between the systems state x, y and
the equilibrium point x∗ , y∗ , which we can designate as x − x∗ , y − y∗ evolves over time. However, similar
to a standard linear system, the sign of the eigenvalues in the exponential functions determine whether
this distance grows or declines exponentially and hence whether the x and y values diverge away from or
converge towards the equilibrium.

v y

y*

x*
u x

Figure 6.4:

Another way of looking at this is by introducing two new variables:

u = x − x∗ v = y − y∗ (6.12)

substituting them in equation 6.11 gives us:

du

dt = au + bv
dv (6.13)
dt = cu + dv.

System 6.13 is a general linear 2D system, but now in the variables u and v. However, u and v are very
closely related to our original variables x and y. u = x − x∗ simply means that u equals x except for a
shift by −x∗ , similarly v = y − y∗ means that v equals y except for a shift by −y∗ . Indeed, in system
6.13 the equilibrium is located at (0, 0) rather than (x∗ , y∗ ). The general solution of 6.13 describes the
behavior around this equilibrium point (u = 0, v = 0). However, as system 6.13 is a linear approximation
of system 6.4 around equilibrium (x∗ , y∗ ) this is also an approximation of the behaviour of the original
system 6.4 around the equilibrium point (x∗ , y∗ ) (Figure 6.4).

Either way, to determine the behaviour of linear system 6.10 or 6.13, we need to determine the val-
ues of the partial derivatives in the equilibrium point, which together constitute the matrix defining the
linearized system:
∂f ∂f
!  
∂x ∂y a b
J= ∂g ∂g = (6.14)
∂x ∂y
c d

this matrix is called the Jacobian of system 6.4 in point (x∗ , y∗ ). It allows us to determine the eigen-
values and hence establish the equilibrium type of both the linearized system and the original system it
approximates.
76 L INEAR APPROXIMATION OF NON - LINEAR 2D SYSTEMS

General plan for phase portrait analysis

In Chapter 4 we discussed the general plan for how to analyse the phase portrait of a system using
the method of finding the systems null-clines and vectorfield. However, this approach is not always
sufficient to find out the type and stability of an equilibrium. However, it is always possible to determine
equilibrium type and stability from the eigenvalues. In case of a linear system these can be directly found
from the matrix of the system, in case of a non-linear system we need to find the Jacobian of the system
in the particular equilibrium point. Thus, this allows an alternative method of phase plane analysis:

1. Find equilibria of the system from dx


dt = 0, dy
dt = 0
2. Find the general Jacobian of the system (if system is non-linear)
3. Find the Jacobian in each equilibrium point, determine eigenvalues and find equilibrium type and
stability
4. Sketch local phase portraits around each equilibrium and connect them to form a global phase
portrait. Distinghuish stable equilibria and their basins of attraction.

6.4 Exercises
0. Read the chapter

(a) What does a partial derivative describe?

(b) How do you linearize an equation?

(c) What is a Jacobian?

(d) Which behaviour of the system does a Jacobian describe?

1. Find partial derivatives of these functions. After finding derivatives evaluate their value at the
given point (if asked).

(a) ∂z
∂x and ∂z
∂y for z(x, y) = x2 + y2 − 4; at x = 1; y = 2

(b) ∂z
∂x for z(x, y) = x(25 − x2 − y2 ); at x = 3; y = 4

(c) ∂z
∂N and ∂z
∂R for z(N, R) = N(bR − d) at R = 0, N = 0; and at R = db , N = 1 :
∂z ∂z a
(d) ∂P and ∂M for z(P, M) = 1+P − bM.

(e) ∂z
∂N and ∂z
∂P for z(N, P) = aN − eN 2 − bNP

(f) ∂z
∂M and ∂z
∂A for z(M, A) = ML − δA − vMA
h+A

∂z ∂z −aP2
(g) ∂P1 and ∂P2 for z(P1 , P2 ) = h+P12 +2P2

∂z ∂z b2 N 2 T
(h) ∂N and ∂T for z(N, T ) = 1+cN+bT N 2

2. Find a linear approximation for the function at the given point.

(a) f (x, y) = x2 + y2 ; at x = 1, y = 1
6.4 E XERCISES 77

3. (A) Find equilibria of the following systems and find the following partial derivatives at each found
equilibrium point ( ∂∂xf , ∂∂yf , ∂g ∂g dx dy
∂x , ∂y ), assuming that dt = f (x, y) and dt = g(x, y).
 dx
(a) dt = −4y
dy
= 4x − x2 − 0.5y
 dt
dx 2
(b) dt = 9x + y
dy
dt = x − y
 dx
(c) dt = 2x − xy
dy
= −y + y2 x
 dt
dx
(d) dt = (1 − x − 3y)x
dy
dt = (1 − 2x − 2y)y

4. Consider the system: dx


dt = x + 4y + ex − 1; dy
dt = −y − y ∗ e
x

(a) Check that (0, 0) is an equilibrium point of the system

(b) Find the general expression for the Jacobian of this system

(c) Find the Jacobian at the point (0, 0)

(d) Write the linearization of the system close to the equilibrium (0, 0)

5. Find the general solution of these systems around their equilibrium point(s). What is the type and
stability of the equilibrium point(s)?
 dx
(a) dt = −2x + y
dy
dt = x − 2y
 dx
(b) dt = 3x − y
dy
dt = −2x + 4y
78 L INEAR APPROXIMATION OF NON - LINEAR 2D SYSTEMS
Chapter 7

Efficient analysis of systems of two


differential equations

In the previous chapters we learned how to determine the type and stability of of an equilibrium from
the original matrix of a linear system or by determining the Jacobian matrix of a non-linear system
in the equilibrium point. From these matrices we computed the eigenvalues to find out which type of
equilibrium the system has. In the next section we will demonstrate an efficient method for determining
the signs and types of the eigenvalues, and hence the type of equilibrium, from the coefficients of the
matrix without actually computing the eigenvalues.

7.1 Determinant-trace method

In this section we derive a simple method for finding signs of eigen values and the type of equilibrium of
a linear or linearized system:
 dx
dt = ax + by (7.1)
dy
dt = cx + dy.

The eigen values of system (7.1) are the roots of the characteristic equation (2.22):

a−λ b
Det = 0, (7.2)
c d −λ

or:

a−λ b
Det = (a − λ)(d − λ) − cb = λ2 − λ(a + d) + ad − cb = 0
c d −λ

Let us express the last equation in a slightly different form using the following definition:

 
a b
Definition 14 The trace of the matrix is trA = a + d.
c d

The determinant of the matrix A is detA = ad − cb


80 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

So we can rewrite the characteristic equation using the definition of the trace and the determinant as
follows:
λ2 − trAλ + detA = 0 (7.3)

We see that although the original system depends on four parameters a, b, c, d the characteristic equation
depends only on two parameters trA and detA. Hence if we know the determinant and the trace of our
system we can find the eigen values and the type of the equilibrium of the system. Indeed, from (7.3) we
can easily find that the eigen values are:

trA ± D
λ1,2 = where D = (trA)2 − 4detA (7.4)
2
Roots of the equation (7.4) are as the roots of any quadratic equation, connected in the following way to
the coefficients of the equation:
λ1 + λ2 = trA (7.5)
λ1 ∗ λ2 = detA (7.6)
To prove the properties (7.5) and (7.6), just note that if λ1 and λ2 are the roots of the quadratic equation
7.3, this equation can be rewritten as: λ2 − trAλ + detA = (λ − λ1 )(λ − λ2 ), which can be rewritten as
λ2 − trAλ + detA = λ2 − (λ1 + λ2 )λ + λ1 λ2 If we now compare the left and the right hand sides of this
last equation we will get both properties (7.5) and (7.6).

Let us start classification.

1. If detA < 0, then D = (trA)2 − 4detA > 0, so we have real roots. From (7.6) we conclude that their
product is negative, i.e. roots have different signs, i.e. λ1 < 0, λ2 > 0, or λ1 > 0, λ2 < 0 and we
have a saddle point.
2. If detA > 0, then D = (trA)2 − 4detA can be negative as well as positive. This means the roots can
be real or complex. Let us consider the case of real roots first, i.e.

D = (trA)2 − 4detA ≥ 0 (7.7)

If (7.7) holds, the roots are real. Next, let us use the property λ1 ∗ λ2 = detA. In the case of
detA > 0, the product of the roots is positive, i.e. the roots have the same sign. They can be
both positive, or both negative. The sign of the roots can be found from the trace of the matrix
(λ1 + λ2 = trA). When trA > 0, then λ1 > 0, and λ2 > 0 and we have a non-stable node. When
trA < 0, λ1 < 0 and λ2 < 0 and we have a stable node. Let us formulate it as a separate case:
3. If detA > 0, D > 0 and trA < 0 the equilibrium is a stable node.
Let us put this information into a graph (fig.7.1). On this graph let us use trA as the x-axis and
detA as the y-axis. Case 1 of a saddle point then corresponds to the lower half plane (region 1).
The line (7.7), which separates real and complex roots, is the parabola given by detA = (trA)2 /4,
or y = x2 /4. Real roots are below this line. Therefore, in region 2, where trA > 0, we have case 2
of a non-stable node. In region 3 trA < 0 and we have case 3 of a stable node.
4. If detA > 0, and D < 0, we have complex roots. In accordance with (7.5) and (??) they are:

trA −D
λ1,2 = ±i
2 2
Hence, Reλ1,2 = trA/2. From this we immediately see that if trA > 0, we have a non-stable spiral
(region 4).
5. If detA > 0, D < 0 and trA < 0, we have a stable spiral (region 5).
7.1 D ETERMINANT- TRACE METHOD 81

det A
D=0
stable spiral non−stable spiral

6
3 5 4 2
stable center non−stable
node node

tr A
1
saddle

Figure 7.1:

6. The last case of a center point appears when Reλ1,2 = trA/2 = 0, or when trA = 0 and detA > 0.
In our graph is it the upper part of the detA axis (region 6)

Let us apply this method for several examples:

Example. Find the equilibrium type for the following systems:


 dx      dx    
dt 1 2 x dt 4 1 x
a) dy = ; b) dy = ;
dt
3 4 y 1 2 y
 dx      dxdt    
dt −2 3 x dt 1 −1 x
c) dy = ; d) dy = ;
dt
−2 1 y dt
2 −1 y

Solution Our plan is to find detA,trA for the corresponding matrices and make a conclusion.
 
1 2
a) The matrix is A = ; trA = 1 + 4 = 5; detA = 1 ∗ 4 − 2 ∗ 3 = −2. Therefore the eigenvalues
3 4
must have opposite signs and we have case 1, so the system has a saddle point.


4 1
b) The corresponding matrix is A = ; trA = 4 + 2 = 6; detA = 4 ∗ 2 − 1 ∗ 1 = 7. As detA > 0
1 2
we need to check that the roots are real. We have D = (trA)2 − 4detA = 6 ∗ 6 − 4 ∗ 7 = 36 − 28 > 0, hence
the roots are real and we have case 2, hence the system has a non-stable node.

c) trA = −2 + 1 = −1; detA = −2 + 6 = 4; D = (trA)2 − 4detA = 1 − 4 ∗ 4 = −15 < 0, we have complex


roots and we are in region 5 and have a stable spiral

d) trA = 1 − 1 = 0; detA = −1 + 2 = 1, so we have case 6 and we have a center point, or oscillation in


our system.
82 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

7.2 Graphical Jacobian

It turns out that in certain cases we do not even have to compute the actual values of the partial derivatives
making up the Jacobian in the equilibrium point, but that information on the sign of the partial derivatives
(+, −, 0) in the equilibrium point is sufficient. This information can be obtained from the vector field
and null-clines of the system.

The main idea behind this method can be seen in Fig.7.2. Let us consider an equilibrium point (x∗ , y∗ ).
To be able to determine the type and stability of the equilibrium we need to determine the Jacobian, and
hence the partial derivatives ∂∂xf , ∂∂yf , ∂g ∂g ∗ ∗
∂x and ∂y in the equilibrium point (x , y ).

In Chapter 1 we learned that ddxf (x, y) ≈ f (x+h,y)−


h
f (x,y)
: the change in f as a function of an increase in x
is the difference in f value between the original point x, y and a nearby point with a slightly higher x
value x + h, y divided by the distance between the two points h. Similarly, ddyf (x, y) ≈ f (x,y+h)− h
f (x,y)
: the
change in f as a function of an increase in y is the difference in f value between the original point x, y
and a nearby point with a slightly higher y value y + h divided by the distance between the two points
h.

As our original point of interest is the equilibrium (x∗ , y∗ ), we will use the point (x∗ + h, y∗ ) (just to
the right of the equilbrium) to determine partial derivatives with respect to x and we will use the point
(x∗ , y∗ + h) (just above the equilibrium) to determine partial derivatives with respect to y. Furthermore
as (x∗ , y∗ ) is an equilibrium point f (x∗ , y∗ ) = 0 and g(x∗ , y∗ ) = 0. Thus we obtain the following approxi-
mation for the Jacobian around the equilibrium point:

f (x∗ +h,y∗ ) f (x∗ ,y∗ +h)


∂f ∂f
!
∂x ≈ h ∂y ≈ h
J= ∂g g(x∗ +h,y∗ ) ∂g f (x∗ ,y∗ +h) (7.8)
∂x ≈ h ∂y ≈ h

Obviously, this approximation will be better if points (x∗ + h, y∗ ) and (x∗ , y∗ + h) are closer to the equilib-
rium point, i.e., if h is small. It turns out that in many cases the exact values of the partial derivative will
not be important for us and we will be able to use the determinant trace method to find the equilibrium
type from just the sign of the components of the Jacobian.

From (7.8) it is clear that the sign of the Jacobian components is the same as the sign of the functions at
the appropriate points, e.g. the sign of ∂∂xf is the same as the sign of f (x∗ + h, y∗ ), etc. The sign of the
functions f (x, y), g(x, y) is represented on the vector field of our system. Indeed, ’→’ means that dxdt > 0
dy
and hence f (x, y) > 0, and ’↑’ means that dt > 0 and hence g(x, y) > 0.

So:
1) in the point (x∗ + h, y∗ ) we can determine partial derivatives with respect to x, whereas in the point
(x∗ , y∗ + h) we can determine partial derivatives with respect to y for the point (x∗ , y∗ ).
2)→/← gives the sign of f (x, y) in that point, whereas ↑/↓ gives the sign of g(x, y) in that point.

If we combine these two facts, we can use the vector field close to the equilibrium point (x∗ , y∗ ) to
determine the sign of the components of the Jacobian matrix in that point.
1) in point (x∗ + h, y∗ ) →/← determines the sign of ∂∂xf
2) in point (x∗ + h, y∗ ) ↑/↓ determines the sign of ∂g
∂x
3) in point (x∗ , y∗ + h) →/← determines the sign of ∂∂yf
4) in point (x∗ , y∗ + h) ↑/↓ determines the sign of ∂g
∂y
7.2 G RAPHICAL JACOBIAN 83

Indeed in figure 7.2 we see that if we look at the horizontal arrows in the point right from the equilibrium
that ∂∂xf = −, whereas if we look at the vertical arrows in this same point we get ∂g
∂x = +. Similarly, if we
look at the horizontal arrows in the point just above the equilibrium we find that ∂∂yf = −, finally if we
∂g
look at the vertical arrows in this same point we obtain ∂y = −.

Figure 7.2: In a we show the nullclines, vector field and location of the equilibrium. In b we show the
two reference points, one located to the right and one located just above the equilibrium, used to compute
the partial derivatives. In c we show the vector field in these two reference points

Three notes on the application of these rules:


1) The testing points must be exactly horizontal (x∗ + h, y∗ ) and exactly vertical (x∗ , y∗ + h) with respect
to the equilibrium point.
2) Testing points should be as close as possible to the equilibrium and should never cross a null cline
when going from the equilibrium.
3) If a null cline is exactly horizontal, a position to the right of the equilibrium will lie on this null cline
and hence ∂g ∂x = 0. If a null cline is exactly vertical, a position just above the equilibrium will lie on this
null cline and hence ∂∂yf = 0.

x’=0 x’=0
y’=0 y y
y
x’=0

y’=0 y’=0

x x x
a b c
Figure 7.3:

Examples. Let us consider several examples of the application of this graphical approach for null-clines
presented in Fig.7.3 and Fig.7.4. On all these figures the testing points are marked by filled circles.
 
−α −β
From fig.7.3a we find the following components of the Jacobian: J = , where α, β, γ, δ
+γ −δ
stand for unknown positive numbers. Thus we see that detJ = αδ + βγ > 0, and trJ = −α − δ < 0, thus
(see fig.7.1) we have a stable equilibrium (stable node or stable spiral). On the basis of these data we
cannot say whether this equilibrium is a node or a spiral as we cannot compute whether the discriminant
of the Jacobian is positive or negative, as this depends on the exact values of the partial derivatives.
 
−α 0
From fig.7.3b we find: J = . Thus detJ = αδ > 0, trJ = −α − δ < 0 and D = tr2 − 4Det =
0 −δ
84 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

(α+δ)2 −4αδ = α2 +2αδ+δ2 −4αδ = α2 −2αδ+δ2 = (α−δ)2 > 0, thus we have a stable node, which
we could also see directly from the vectorfield.
 
+α 0
From fig.7.3c we find: J = . Thus detJ = −αδ < 0, and we have a saddle, which we could
0 −δ
also derive from the vectorfield.

Figure 7.4:
 
+α +β
From fig.7.4a we find: J = , hence detJ = −αδ − γβ < 0, and we again have a saddle.
+γ −δ
 
+α +β
From fig.7.4b we find: J = . detJ = −αδ + γβ, trJ = α − δ. Because we do not know
−γ −δ
the values of the coefficients α, β, γ, δ, just their signs, we do not know if the detJ and trJ are positive or
negative, thus we cannot determine the equilibrium type in this case using the graphical Jacobian.
 
+α +β
Finally from 7.4c we find: J = . Thus detJ = αδ − γβ, trJ = α + δ > 0. Again we cannot
+γ +δ
determine the sign of the detJ, but positive trJ implies that the equilibrium will be unstable. We do not
know its type: non-stable node, non-stable spiral and a saddle are all possible.

We see that the method of graphical Jacobian is a useful tool for finding the equilibrium type from the
vector field, but sometimes it is not sufficient and we need to know not only the sign but also the values
of the Jacobian coefficients. In this case we need to compute the values of ∂∂xf , ∂∂yf , ∂g ∂g
∂x and ∂y in the
equilibrium point.

7.3 Plan of qualitative analysis

Let us now formulate a plan to qualitatively study systems of two differential equations and consider
several examples. Our main aim is to plot the phase portrait and determine equilibria types and stability,
and basins of attraction to be able to predict the dynamics of the system. Based on the methods which
we have developed in the previous chapters we arrive at the following general approach:

Global plan

Null cline and vector field approach


dy
1. Draw the dx
dt = 0 null-clines from the equation f (x, y) = 0 and the dt = 0 null-clines from the
equation g(x, y) = 0 using different colors.
7.3 P LAN OF QUALITATIVE ANALYSIS 85

2. Choose a point in an “extreme”(variables are big or small there) region on the x, y plane and find
the vector field for the x-component. Denote the x component as a ’→’ if f (x, y) > 0 and as a ’←’
if f (x, y) < 0. Use the same color as you used for the x null cline.
3. Find the vector field for the y-component at the same point. Denote the y component as a ↑ if
g(x, y) > 0 and as a ↓ if g(x, y) < 0. Again use the same color as you used to draw the y null cline.
4. Find the vector field in the adjacent regions using the following rules:

(a) change the direction of the x (horizontal) component of the vector field if to get to the
adjacent region you cross the x null-cline (use the matching colors).

(b) change the direction of the y (vertical) component of the vector field if to get to the adjacent
region you cross the y null-cline (use the matching colors).

(c) show the direction of the vector field on the null-clines close to the equilibrium

5. Look at the vector field in the four different regions surrounding an equilibrium point to see if this
provides enough information on the type and stability of the equilibrium.

Graphical Jacobian approach


If the vector field does not allow you to determine the stability of the equilibrium, determine the graphical
Jacobian of the equilibrium point using the following approach:

1. For each equilibrium point (point of intersection of different null-clines) choose two points, one of
which is located to the right and the other upward from the equilibrium. Find the components of
the Jacobian using the following rules:

(a) the sign of the horizontal component (x) of the vector field in the point to the right of the
equilibrium point gives the sign of ∂∂xf .

(b) the sign of the vertical component (y) of the vector field in the point to the right of the
equilibrium point gives the sign of ∂g
∂x .

(c) the sign of the horizontal component (x) of the vector field in the point upward of the equi-
librium point gives the sign of ∂∂yf .

(d) the sign of the vertical component (y) of the vector field in the point upward of the equilib-
rium point gives the sign of ∂g
∂y

2. Try to compute the sign of the determinant and of the trace of the Jacobian and try to identify the
type of equilibrium from fig.7.1.

Analytical Jacobian approach


If the stability of the equilibrium can not be determined from the vector field or the graphical Jacobian,
we need to determine the exact values of the components of the Jacobian using the following approach:

1. Find the equilibria of your system from the equations f (x, y) = 0 and g(x, y) = 0.
2. For each equilibrium point (x∗ , y∗ ), find the Jacobian at that equilibrium point
 
∂ f /∂x ∂ f /∂y
J= (7.9)
∂g/∂x ∂g/∂y (x∗ ,y∗ )

Note: Do not forget to substitute x = x∗ , y = y∗ into the Jacobian.


86 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

3. Determine the type of each equilibrium (x∗ , y∗ ). For this compute


 
∂ f ∂g ∂ f ∂g
detJ = ∗ − ∗ (7.10)
∂x ∂y ∂y ∂x (x∗ ,y∗ )
 
∂ f ∂g
trJ = + (7.11)
∂x ∂y (x∗ ,y∗ )

D = (trJ)2 − 4detJ (7.12)


To find the type of equilibrium use fig.7.1

Independent of how you determined the type and stability of the equilibria of the system, we next do the
following:

1. Based on the obtained knowledge of the type and stability of the equilibrium and the vector field
and nullclines, draw a local phase portrait around each equilibrium point.
2. Connect the different local phase portraits into a global phase portrait and determine and indicate
attractors and their basins of attraction.

Example Study the following system (Lotka-Volterra model):


 dx
dt = 3x(1 − x) − 1.5xy (7.13)
dy
dt = 0.5xy − 0.25y

Solution We have studied different aspects of this model throughout the reader, so let us collect the
information we already have.

I. The null-clines and vector field of this system were given in fig.4.7. We repeat them in fig7.5a: We see

y y

x x
0 0.5 1 1 0.5 1

Figure 7.5:

three equilibria, one at (0,0), one at (0.5,1) and one at (1,0). From the vector field we see that both (0,0)
and (1,0) have one stable and one unstable direction, so these must be saddle nodes and hence unstable.
For equilibrium (0.5,1) the type can not be found from the vector field.

II. For the graphical Jacobian


 we obtain
 for the three equilibria:
+α 0
For point (0, 0), J = , detJ = −αδ < 0, thus we have a saddle.
0 −δ
 
−α −β
For point (1, 0), J = , detJ = −αδ < 0, thus we have a saddle.
0 +δ
7.4 E XERCISES 87

 
−α −β
For point (0.5, 1),J = , detJ = γβ > 0, trJ = −α < 0, thus we have a stable equilibrium
+γ 0
(stable node or stable spiral).

III. Next let us do an analytical Jacobian analysis:


In Chapter 4 we found three equilibria points: (0, 0), (1, 0) and (0.5, 1). 
3 − 6x − 1.5y −1.5x
We compute the general Jacobian as:: J = .
0.5y 0.5x − 0.25
Let us find all equilibria
 types from  this Jacobian.
3 0
At point (0, 0) J1 = , detJ1 = 3 ∗ (−0.25) < 0, thus this is a saddle point.
0 −0.25
 
−3 −1.5
At the point (1, 0) J2 = , detJ2 = (−3) ∗ 0.25 < 0, thus this is also a saddle point.
0 0.25
 
−1.5 −0.75
At the point (0.5, 1) J3 = , detJ3 = (−1.5) ∗ 0 − 0.5 ∗ (−0.75) < 0.375, thus we need
0.5 0
to find trJ3 = −1.5 and D = (−1.5)2 − 4 ∗ 0.375 = 0.75 > 0 thus this is a stable node.

Now that we know all equilibria types we can draw the local phase portraits around them: fig.7.5b. From
the null-cline analysis we can find the approximate locations of the manifolds of the saddle points and
the direction of the trajectories around the stable node.

Finally, we can draw the global phase portrait. There are no general rules to do this. However in the case
of fig.7.5b we can expect that the non-stable manifold of the saddle point at (0, 0) will end at the other
saddle point (1, 0), the non-stable manifold of the saddle point (1, 0) as well as most of other trajectories
should go to the stable node at (0.5, 1) as this is the only attractor here. Thus we get the phase portrait
presented in fig.7.5c. We see that we will have a stable global attractor with as a basin of attraction the
whole region x > 0, y > 0 (except two axes x = 0; y = 0).

We see that the phase portrait which we got as a result of our study is qualitatively the same as the phase
portrait of this system obtained using a computer (see fig.??b). Note that if we only use the graphical
and not the analytical Jacobian, the only difference was that were not able to determine that the stable
middle equilibrium is a node. However, this has almost no consequences for the phase portrait.

7.4 Exercises
0. Read the chapter

(a) What is a determinant?

(b) What is a trace?

(c) How is the determinant related to the eigenvalues?

(d) How is the trace related to the eigenvalues?

(e) How can you determine equilibrium type from determinant and trace?

(f) What is a graphical Jacobian?

1. Findthe type and stability ofthe equilibria of the below


 dx systems using the
 det-tr method.
dx dx dx
(a) dt = 3x + y (b) dt = 2x + y (c) dt = 2x + y (d) dt = x + 10y
dy dy dy dy
dt = −20x + 6y dt = 2x − 10y dt = 5x − 2y dt = −10x − y
88 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS

2. 
Consider the following model for interacting algea (x) and zooplankton (y) populations:
dx
dt = 2x(1 − y) x ≥ 0;
dy 2
dt = 2 − y − x y ≥ 0.

(a) Find all equilibria of the system.

(b) Find the general expression for the Jacobian of this system.

(c) Determine the type of each equilibrium using the det-tr method.

(d) Draw qualitative local phase portraits around each equilibrium point and connect them into
a global phase portrait.

3. Now study the algea-zooplankton model of the previous question again using the graphical Jaco-
bian approach:

(a) Sketch the vector field for the system using null-clines.

(b) Find type and stability of equilibria using the graphical Jacobian.

(c) Compare your results to the previous question.

4. Complete the vector field approximations for the null-clines shown in fig7.6a,b,c. Mark equilibria
and determine their type using the graphical Jacobian method. Draw local and then global phase
portraits. Try to describe the dynamics of the system, by saying what happens with the variables x
and y in the course of time.

y y y

x x x
a c
b
Figure 7.6:

dN

= aN − bNP
dt
5. Study the Lotka-Volterra model for a predator-prey system: dP
= cNP − dP
dt
here N denotes the prey population, P denotes the predator population and a > 0, b > 0, c > 0, d > 0
are parameters

(a) Find the nontrivial equilibrium of the system (i.e. an equilibrium where N 6= 0, P 6= 0).

(b) Find the linearization of the system at this point (i.e. the general Jacobian matrix)

(c) Determine the type and stability of this equilibrium

(d) Sketch the phase portrait around this equilibrium. Which kind of dynamics do we expect
here?

6. Now study the Lotka Volterra model of the previous question again using the graphical Jacobian
approach:

(a) Sketch the vector field for the system using null-clines.
7.4 E XERCISES 89

(b) Find type and stability of equilibria using the graphical Jacobian. Sketch local phase por-
traits.

(c) Compare your results to the previous question.

(d) Does the vector field, Jacobian and equilibrium types change if we change the values of
parameters a, b, c, d? How?

7. Sketch the phase portraits of the following systems using null-clines. Try to find the type of
equilibrium using the graphical Jacobian. If you are unable to find the equilibrium type in this way
Jacobian, find it using the trace-determinant method. Sketch a qualitative phase portrait.
 dx
(a) dt = 3x + y
dy
dt = −x − y
 dx
(b) dt = x + 2y
dy
dt = −2x − 2y
 dx
(c) dt = y
dy
dt = 3x
 dx
(d) dt = x − 4y
dy
dt = x + y
90 E FFICIENT ANALYSIS OF SYSTEMS OF TWO DIFFERENTIAL EQUATIONS
Chapter 8

Limit cycle

8.1 Stable and non-stable limit cycles

In previous chapters we found several possible types of equilibria: saddle, node, spiral and center. Stable
nodes and stable spirals are stable equilibria, or attractors, which determine the ultimate dynamics of a
system. Note that these attractors correspond to a single point on the phase plane. As a consequence,
once the system has converged to these attractors, the system will stay at this single point in phase space
and the variables will hence maintain constant values.

However, it turns out that there exists another important type of attractor for systems of two (or more)
differential equations. It is called a limit cycle. On the phase plane, a limit cycle appears as a closed
curve connecting nearby points into a closed loop (fig.8.1). If a limit cycle is stable it is as an attractor.
In contrast to stable nodes and stable spirals, a stable limit cycle is not a single point on the phase plane
but rather a closed curve of points on the phase plane. As a consequence, once the system has converged
to the limit cycle, it will perpetually walk along the curve of the limit cycle, causing the variables of
the system to continuously change in a cyclical repetetive manner, rather than staying at a single point.
However, unless perturbed, the system will not leave the curve and the change in variables is hence
constrained to this curve.

Figure 8.1a shows a stable limit cycle (bold ellipse). The figure also shows two trajectories, one starting
inside and one starting outside the limit cycle. As the limit cycle is stable, it acts as an attractor and
the inside trajectory will spiral outward towards the limit cycle whereas the outside trajectory will rotate
inward towards the limit cycle. The dynamics inside the stable limit cycle show that there is an unstable
spiral inside the limit cycle. Indeed, there must always be at least one equilibrium inside the limit cycle.

Figure 8.1b shows an unstable limit cycle (dashed ellipse). Again, the figure also shows two trajectories,
one starting inside and one outside the limit cycle. As the limit cycle is unstable, the trajectories diverge
away from the limit cycle. As a consequence the trajectory inside the limit cycle spirals inward, consis-
tent with the presence of a stable spiral that acts as an attractor inside the unstable limit cycle. Where the
trajectory outside the limit cycle goes depends on the rest of the system, if there is an attractor outside
the limit cycle it will go there, otherwise the system will go to plus or minus infinity.

The main questions regarding limit cycles are: what will be the dynamics of systems with limit cycles
and how do limit cycles occur in systems of two differential equations? We will also consider an example
of a biological system with a limit cycle.
92 L IMIT CYCLE

y B y

C A C
A
B
x x
a b
Figure 8.1: Phase portrait of a system of two differential equations with either a stable (a) or unstable (b)
limit cycle.
x
x x A

C
B
time time time
a b c

Figure 8.2: Dynamics of a system with a stable limit cycle (fig.8.1a).

x x A x
B
C

time time time


a b c

Figure 8.3: Dynamics of a system with an non-stable limit cycle (fig.8.1b).

8.2 Dynamics of a system with a limit cycle.

What will be the dynamics of systems with a limit cycle? To get a better understanding of this, let us
first consider the equilibrium type called “center” that we studied in section 5.2. The phase portrait of
that point (fig.5.6) was a set of embedded ellipses. The corresponding dynamics are continuous, constant
amplitude oscillations of the x and y variables. As a center is neither attractive nor repulsive but exactly
between these two behaviours, the initial conditions will determine on which of the set of embedded
ellipses the system resides and hence what will be the amplitude with which x and y oscillate. Due to
the “neutrality” neither convergence to smaller amplitude oscillations (smaller ellipse) nor divergence to
larger amplitude oscillations (larger ellipse) will occur.

A limit cycle is just a single ellipse. A trajectory following the limit cycle will thus also correspond to
oscillations of the x and y variable. However, as there is only a single ellipse, only oscillations of one
particular amplitude occur. Figures 8.2a and 8.3a show the dynamics of the variable x for trajectories
starting exactly at the limit cycles shows in figure 8.2a and 8.2b respectively. The dynamics of trajectories
starting close to a limit cycle will depend on whether the limit cycle is stable or not. If the limit cycle is
stable, then a trajectory which starts inside the limit cycle will approach it in the course of time and we
obtain oscillations with initially increasing amplitude (fig.8.2b). If the trajectory starts outside the limit
cycle, then we obtain oscillations with an initial decrease of amplitude until the trajectory reaches the
limit cycle (fig.8.2c).
8.3 H OW DO LIMIT CYCLES OCCUR ? 93

For a non-stable limit cycle the dynamics are different. Only the trajectory which starts on a limit cycle
will have oscillatory dynamics (fig.8.3a). Other trajectories will show different behavior. The trajectory
which starts inside the limit cycle will approach the stable equilibrium inside the limit cycle and we
obtain oscillations with gradually decreasing amplitude (fig.8.3b). The trajectory which starts outside
the limit cycle blows up to infinity or converges to another equilibrium (fig.8.3c). The oscillations shown
in fig.8.3a are highly improbable in real systems. This is because for such oscillations the trajectory must
start exactly on the limit cycle and even small disturbances will switch our system either to the behavior
of fig.8.3b or fig.8.3c. Therefore, for real systems the non-stable limit cycle just determines the basin of
attraction of the stable equilibrium which is located inside this limit cycle.

8.3 How do limit cycles occur?

In many cases the limit cycle occurs as a result of the changing of a parameter in a system of differential
equations. The most usual process of formation of a limit cycle is the following. Assume we have a
system of two differential equations with a parameter c:
 dx
dt = f (x, y, c)
dy (8.1)
dt = g(x, y, c)

Assume that at some parameter value c = c1 system (8.1) has a global attractor which is a stable spiral
(fig.8.4a). This means that all the trajectories which originate close to or even far from this equilibrium
approach it in the course of time. Because our equilibrium point is a stable spiral the Jacobian of the
system at this point will have two complex eigen values λ1,2 = α ± iβ; α < 0. However, because our
system will now depend on the parameter c the eigen values will also depend on this parameter:

λ1,2 (c) = α(c) ± iβ(c); (8.2)


Because we have assumed that at c = c1 system (8.1) has an equilibrium which is a stable spiral, then
α(c1 ) < 0. When we change the parameter c the value of α(c) will change and at some c2 it can become
a positive number α(c2 ) > 0. This means that the stable spiral in fig.8.4a will become an unstable spiral.

Now recall that we are studying a non-linear system here. The eigenvalues of the linearization of that
system around an equilibrium only tell us something about the dynamics of the original system close
to the equilibrium but do not tell us anything about the global dynamics. If we want to now about the
global dynamics we fill in (x, y) points that lie far away from the equilibrium point. However, as the
dy
equations of the system of course have not changed, we will get the same dx dt = f (x, y) and dt = f (x, y)
directions of the vectorfield as before. So the equations tell us that globally the direction of the dynamics
has not changed and we still have a convergent flow towards the equilibrium, whereas the eigenvalues
tell us that locally -close to the equilibrium- the dynamics has changed and now diverges away from
the equilibrium (fig.8.4b). Hence, in our phase portrait we have two types of flow: the diverging flow
from the unstable spiral around the equilibrium and the converging flow from the periphery. Simple
geometrical consideration shows that these two flows must be separated from each other. The line of
separation will be the limit cycle in our system. Therefore, we will obtain a phase portrait as in fig.8.1a.

Note, that such a mechanism of limit cycle formation frequently occurs in systems of two equations. It
is called the Hopf bifurcation. From our analysis it follows that the Hopf bifurcation occurs when α(c)
changes its sign, i.e. at the parameter value c∗ where:

α(c∗ ) = 0 (8.3)
94 L IMIT CYCLE

y y

x x
a b

Figure 8.4: Appearance of a limit cycle in a system with a parameter.

Note that at this value for c the equilibrium is between a stable and unstable spiral, and hence is neither
an attractor nor a repellor. Indeed, we know that for λ1,2 = α ± iβ with α = 0 we have a center point.

8.4 Example of a system with a limit cycle

Let us consider the Holling-Tanner model for predator-prey interactions.


aRP
dP/dt = rP(1 − KP ) − d+P

R (8.4)
dR/dt = bR(1 − P ) P > 0; R > 0

here P denotes the prey and R the predator population. The term rP(1 − P/K) describes the logistic
aRP
growth of the prey in absence of the predator, and d+P accounts for the amount of prey eaten by the
predator which encorporates a saturating functional response. Finally, R(1 − RP ) represents the logistic
growth of the predator with a carrying capacity that is assumed to depend linearly on the size of the prey
population. At a = 1, b = 0.2, r = 1., d = 1., K = 0.7 system (8.4) has a single equilibrium point (for
P > 0, R > 0) which is a stable spiral. The null-clines, phase portrait and dynamics of this system are
shown in fig.8.5.
r r
10

10 10

5 5 5

p
r

-4 -4
10 10
-4 -4 0
10 5 10 p 10 5 10 p
0 37.5 75 112 150

Figure 8.5: Dynamics of system (8.4) for K = 0.7. a-null-clines; b-an orbit; c-time-plot for the both
variables for the orbit from fig.b

If we increase the value of the parameter K which accounts for the carrying capacity (K = 1.0) the type
of equilibrium changes and it becomes an unstable spiral. As we discussed in section 8.3 we expect the
formation of a stable limit cycle. We can clearly see it in fig.8.6b. The trajectory, which starts from the
same initial conditions as the trajectory from fig.8.5b will now approach some closed curve which is a
limit cycle and the dynamics of the system will be oscillatory (fig.8.6c).
8.5 E XERCISES 95

r r
10

10 10

5 5 5

-4 -4 p
10 10
-4 -4 0
10 5 10 p 10 5 10 p
0 37.5 75 112 150

Figure 8.6: Dynamics of system (8.4) for K = 1.0. a-null-clines; b-an orbit; c-time-plot for both variables
for the orbit from fig.b
r r
10

10 10

p
5 5 5

-4 -4
10 10
-4 0 -4
10 5 10 p 10 5 10 p
0 37.5 75 112 150

Figure 8.7: Dynamics of system (8.4) for K = 1.0. a-an orbit originating inside the limit cycle; b-time-
plot for the both variables for the orbit from fig.a; c-phase portrait of system (8.4)

If we start a trajectory inside the limit cycle then, as we predicted in fig.8.1a and fig.8.2b, the trajectory
will approach the limit cycle and the dynamics will be oscillation with increasing amplitude fig.8.7a,b.
The complete phase portrait of this system at K = 1 is shown in fig.8.7c.

Conclusion 2 A limit cycle is a closed trajectory on the phase portrait of a system of two differential
equations.

If trajectories around the limit cycle converge onto it, then the limit cycle is called a stable limit cycle. If
trajectories around the limit cycle diverge away from it, then the limit cycle is called a non-stable limit
cycle.

The dynamics of a system with a stable limit cycle is oscillatory. The dynamics of a system with an
non-stable limit cycle is either converging to the equilibrium which is located within the limit cycle or
diverging, possibly to infinity.

Limit cycles can appear as a result of a Hopf bifurcation, i.e. the process where the real part of the
complex eigenvalues change their sign.

8.5 Exercises
0. Read the chapter

(a) What is a limit cycle?

(b) What are oscillations?


96 L IMIT CYCLE

(c) When do systems with limitcycles show oscillatory dynamics?


1. Assume that a system of two differential equations has two equilibria which are a saddle point and
a non-stable spiral. The phase portrait of this system is partially shown in fig.8.8a
(a) What is missing here?: Complete the phase portrait of this system

(b) Qualitatively sketch the dynamics of the variable x (dependence of the variable x on time)
for the initial conditions which are shown in fig.8.8 by the point A in fig.8.8a.
2. Complete the phase portrait of the two systems shown in fig.8.8b and c. Qualitatively sketch the
dynamics of the variable x ( dependence of the variable x on time) for the initial conditions which
are shown by the points A,B,C in fig.8.8b and c. (Note, that in fig.8.8b and c a stable limit cycle is
drawn with a solid line and a non-stable limit cycle is drawn with a dashed line)

y − y −B y − B
3− 3− 3−
− A −
C A −
C A
1−
1− 1−
| | | | | |
1 3 5 x | | | | | | | | | | | |
1 3 5 x 1 4 7 x
a b c
Figure 8.8:

3. As we discussed in (8.3) the necessary condition for the appearance of a limit cycle via a Hopf
bifurcation is that the eigenvalues of the Jacobian matrix of a system at the equilibrium point are
complex and the real part of the eigenvalues are zero α(c∗ ) = 0. Prove that this condition can be
rewritten in the following way using the det and tr of the Jacobian matrix:

trJ(c∗ ) = 0; detJ(c∗ ) > 0 (8.5)

4. One of the classical models for oscillatory phenomena in biochemical systems is a model called
the Brusselator. In dimensionless form this model can be written as the following system of two
differential equations:
 dx 2
dt = a − (b + 1)x + x y x > 0; y > 0
(8.6)
dy 2
dt = bx − x y a > 0; b > 0
here x and y are concentrations of two biochemical species and a and b are parameters. Study
system (8.6) for a = 1.
(a) Find a non-trivial equilibrium.

(b) Determine the stability of this equilibrium as a function of the parameter b.

(c) Find the value of b when system (8.6) undergoes a Hopf bifurcation. (Note: it can be helpful
to use equations (8.5)).

(d) Assume that the system has stable oscillations of a single constant amplitude. For which
values of b do you expect these oscillations in system (8.6)?
5. Study the following predator-prey model

dP/dt = rP(1 − KP ) − h+P


aRP

(8.7)
dR/dt = caRP
h+P − dR P ≥ 0; R ≥ 0
8.5 E XERCISES 97

(a) Draw null-clines of this system for r = 1, a = 3, h = 0.1, c = 1., d = 2.5 and two values of
K, K = 0.8 and K = 1.6. (Hint: the maximum of the parabola A(x − a)(x − b) is reached at
the middle between its roots (i.e. at x = a+b
2 )).

(b) Determine the stability of non-trivial equilibrium in both cases using the graphical Jacobian.

(c) For which value of K do you expect oscillatory behavior?

(d) Extend your analysis for arbitrary positive values of the parameters (r, a, h, c, d, K > 0) pro-
vided ca > d. Find for which values of K the non-trivial equilibrium is stable. When do
we expect oscillations? (Note, that critical values of K should be a function of the other
parameters of the system).
98 L IMIT CYCLE
Chapter 9

Extra Questions

Chapter 1
1. Perform the indicated operations:

(a) ((x − 2y) ∗ (y − 2x) + 2y2 ) ∗ 1x


a−2b 4b−2a
(b) 2p : √
p

2. 
Solve the system of equations for the specified variables and find A, B,C in:
 rA(1 − KA − AB) = 0
AB − dB − BC = 0 r, K, d, f > 0
BC − fC = 0

3. Find the derivative of f (x):

(a) f (x) = 2x
q
(b) f (x) = x13

(c) f (g) = cos(x2 );

(d) f = (cos(x))2 ;

(e) y = ax ∗ ebx a, b > 0


x2 −5
(f) y = 2x2 −3x

ax2
(g) y = bx−c , a, b, c > 0
x
(h) y = 1+ dx , d>0

xn
(i) y = xn +an , a>0

4. Solutions of differential equations

(a) Show that function N(t) = A ∗ e3t , where A is an arbitrary constant, is a solution of the
differential equation: dN(t)
dt = 3N. For that, compute the derivative of N(t) substitute this
derivative and the function itself int to the equation dN(t)
dt = 3N and show that the left hand
side of the equation equals to the right hand side.
100 E XTRA Q UESTIONS

(b) Show, using the same steps as before that the function N(t) = s(1 − et ) + Ae−t , where a is an
arbitrary constant and s is a parameters, is a solution of the differential equation dN(t)
dt = s−N

5. Assume that x(t) is an unknown function of t. For f (x) listed below find the following derivatives:
df df
dx and dt .

(a) f (x) = x3

(b) f (x) = e−ax


df
(c) Find the expression for dt for an arbitrary f (x(t)).

6. Without plotting the function find the following information about their graph: find the y-intercept
and zeros; find horizontal or/and vertical asymptotes (if they exist).
x−4
(a) function y(x) given by y = x2 −3x+2
b
(b) function y(x) given by y = a : x3 −c
a, b, c > 0

7. Sketch graphs of the following functions:

(a) x = 4e−3t

(b) y = x2 + 2x − 3
2
(c) y = x+3
2
(d) y = x2bx+a2 + 4. Find how the shape of the graph depends on the value of the parameters a > 0
and b > 0.
2
(e) y = x3bx+c3 , Find how the shape of the graph depends on the value of the parameter c > 0 and
b > 0.

(f) f (n) = rn ∗ (1 − nk ) − h, k 6= 0 find for which values of the parameter h > 0 the graph touches
the n-axis. Tip: draw graph for h = 0 and think about how h affects this graph.

8. Sketch qualitative graphs of the following pairs of implicit functions on the same graph. Find all
intersection points.

(a) aN − P(1 + eP) − bP = 0 and bP − cN = 0 parameters a, b, c, e > 0


NR NR
(b) rR(1 − R/K) − h+R = 0 and h+R − dN = 0 parameters r, K, h, d > 0

Chapter 2
1. Perform the indicated operations:

(a) 32 − 90

(b) (−1 + 2i) + (4 + 7i)

(c) (4 + 5i)(7 + 2i)


1
(d) i

2. Find all roots of the given equations

(a) x2 + 121 = 0
101

(b) x2 + 2x + 3 = 0
3. Find eigen values and eigen vectors of the following matrices:
 
−1 6
(a)
2 −2
 
2 1
(b)
7 −4

Chapter 3
1. Assume that the change in mass of a certain animal can be described using the following differen-
tial equation:
dW
= 400 − 0.3W
dt
where W is the weight in grams and t the time in weeks.
(a) Find the solution for the initial value problem W (0) = 10.

(b) At what time the mass will reach half of the saturated steady state value

(c) If we assume that the linear size of the animal (its length l) is proportional to the cubic root
of the mass (W ), find at what time the object will reach half of its saturated linear size.
2. The dynamics of the ionic channels in the famous Hodgkin-Huxley model for a nerve cell are
described using the following type of equations:
dm
= α(1 − m) − βm
dt
where m is a gating variable and α, β are the parameters.
Find the steady state values of the gating variable m and the characteristic time of approaching this
steady state.
3. Consider a model where the harvesting (hn) is proportional to the sizeof the population:

dn/dt = rn(1 − n/k) − hn (9.1)


Find the maximal yield.
4. Compare the harvesting strategies (3.19) and (9.1). Which strategy is better. Why?

Chapter 4
1. Draw the phase portrait of the following systems of differential equations. Explain their dynamics.
 dx
(a) dt = 2y
dy 2
dt = x − x − 0.5y
 dx 2
(b) dt = x + y
dy
dt = x + y
 dx 2 2
(c) dt = x(25 − x − y ) x ≥ 0; y ≥ 0
dy
dt = y(x − 3)
102 E XTRA Q UESTIONS

 dx
dt = xy
(d) dy − ∞ < x < ∞; −∞ < y < ∞
dt = 4 − y − x2
2. Draw the phase portrait of the following system of differential equations. Explain its dynamics
and whether it is different for different parameter settings: Lotka Volterra model with competition
in the prey population:
dN/dt = aN − eN 2 − bNP


dP/dt = cNP − dP
for a = 3, b = 1.5, c = 0.5, d = 0.25, e = 3.
3. A swimming pool is infested with algae whose population is N(t). The owner attempts to control
the infestation with an algicidal chemical, poured into the pool at a constant rate. In the absence
of algae, the chemical decays naturally; when algae are present it is metabolized by them and kills
them. The equations of the rates of change of N(t) and the concentration of the chemical in the
pool, C(t), are 
dN/dt = aN − bNC
dC/dt = Q − αC − βNC
where a, b, Q, α, β are positive constants. Discuss the meaning of each term in these equations.

(a) Put a = 1, b = 1, α = 1, β = 1 and show, that the system has two non-negative equilibria if
Q > 1, find them, draw the phase portrait of this system and explain the dynamics.

(b) What happens if Q < 1?

(c) * Find the conditions for control of the infestation for arbitrary positive values of a, b, Q, α, β.

Chapter 5
1. Find the vector fields of these systems using null-clines. Find equilibria. Determine the type of
each equilibrium using eigenvalues and draw qualitative local phase portraits.
 dx
(a) dt = x − 3y
dy
dt = x + y
 dx
(b) dt = y
dy
dt = −x
 dx 2
(c) dt = 9x + y
dy
dt = x − y
 dx 2 2
(d) dt = y − x
dy
dt = y − 1

Chapter 6
1. Consider a modification
 dN of the Lotka-Volterra model, which includes competition in the prey popu-
2
lation (−eN 2 ): dt = aN − eN − bNP , where the parameters a, b, c, d, e > 0 and the variables
dP
dt = cNP − dP
N ≥ 0, P ≥ 0.

(a) Find all equilibria of this system.


103

(b) Compute the Jacobian at each equilibrium point.

(c) At which parameter values do we have a non-trivial equilibrium (i.e. an equilibrium at which
both N and P are positive). Find stability of this equilibrium.

2. Consider the following model for cardiac tissue:



de/dt = −e(e − a)(e − 1) − g 0<a<1
(9.2)
dg/dt = εe ε>0

Here the variable e accounts for the transmembrane potential, the variable g accounts for the re-
fractory period and a, ε are the parameters. The shape of the action potential in cardiac tissue is
an important characteristic of myocardium. If the recovery of the transmembrane potential shows
oscillation as in fig.9.1b. It can cause dangerous cardiac arrhythmias. From a mathematical point
of view the oscillations in fig9.1b occur when system (9.2) has an equilibrium point which is a
stable spiral. Monotonous recovery occurs when this equilibrium is a stable node. Determine for

a b

Figure 9.1: The monotonous (a) and oscillatory recovery (b) in an excitable medium

which parameter values we will have situation fig.9.1a and for which parameter values we will
have situation fig.9.1b.
3. The following equations describe the dynamics of predator (P) and prey (N) populations:

 dN/dt = rN(1 − N

K ) − bNP N≥0 P≥0

dP/dt = bNP − 2bP b≥0 K≥0 r≥0

here K, b, r are parameters. This system of differential equations always has an equilibrium corre-
sponding to an extinct population of the predator and non-zero population of prey (P = 0; N 6= 0).

(a) Find this (P = 0; N 6= 0) equilibrium.

(b) Find the Jacobian at this equilibrium point.

(c) For which values of the parameters the predator population can be driven to extinction (i.e.
to that equilibrium ).

(N.B. Do not use the ’graphical Jacobian approach for this problem!).

Chapter 7
1. Find the equilibria of the following systems. Compute the Jacobian at the equilibria points. Deter-
mine type of each equilibrium using det-tr method. Draw qualitative local phase portraits around
each
 equilibrium point.
dx 2
 dx
dt = y − 3x + 2 dt = y
(a) dy 2 − y2 (b) dy 3
dt = x dt = −x + x
104 E XTRA Q UESTIONS

2. The figure below shows the null-clines of three systems of two non-linear differential equations

y y y

x x
x

a b c
(a) Complete the figures by showing the direction of the vector field in all regions on the plane
and on the null-clines.

(b) Mark the equilibria and find their types using the graphical Jacobian (if possible).

3. In theoretical biology the following model has been used to study mRNA (M) protein (P) interac-
tion: (
dP/dt = bM − dP P a, b, d > 0
K2b (9.3)
dM/dt = a K 2 +P2 − dM M P > 0; M > 0
Study this system using the graphical Jacobian approach, i.e. draw null-clines, mark equilibria
as points of intersection of the null-clines, determine stability of these equilibria and sketch a
qualitative phase portrait.
Chapter 10

Dictionary

absolute value absolute waarde


autonomous system autonoom systeem
attractor attractor
basin of attraction basin van attractie
bifurcation bifurcatie
carrying capacity draagkracht
center point centrumpunt
complex conjugate number complex toegevoegde
component of the vector component van de vector
determinant determinant
derivative afgeleide
differential equation differentiaal vergelijking
direction field vectorveld
eigen value eigenwaarde
eigen vector eigenvector
equilibrium evenwicht
general solution algemene oplossing
harvesting oogsten
imaginary part of complex number het imaginaire deel van een complex getal
initial value problem beginwaarde probleem
Jacobian Jacobi-matrix
linear approximation lineaire benadering
modulus modulus
node knooppunt
non-stable manifold instabiele manifold
106 D ICTIONARY

null-cline isocline
parameter parameter
partial derivative partiële afgeleide
particular solution specifieke oplossing
phase space faseruimte
phase portrait faseplaatje
real part of complex number het reëele deel van een complex getal
saddle zadelpunt
spiral spiraalpunt
stability stabiliteit
stable manifold stabiele manifold
system of differential equations stelsel differentiaal vergelijkingen
trajectory trajectorie
trace of the matrix spoor van de matrix
variable variabele
vector vector
Chapter 11

Answers for exercises

Exercises Chapter 1
1. (a) 3axy − 4abx2 − 6by2 + 8b2 xy + 4abx2 + 6by2 − 8xyb2 = 3axy
6(30r+5) 5r2 6(6r+1) r 2 −r2 +36r+6
(b) r(30r+5) − r(30r+5) = r(6r+1) − r(6r+1) = r(6r+1)
2
(a/x )+(q/x ) 2
2. (a) limx→∞ cax+q
2 +x2 = limx→∞ (c2 /x2 )+1 =0
2
aN 2 +q )
(b) limN→∞ b 2 2 = limN→∞ (b/Na+(q/N
3 )+(c2 /N 2 )+d =
a
d
N +c +dN

3. (a) 3r + 2 − 5r − 5 = 6r + 4; so −2r − 3 = 6r + 4; so −7 = 8r; so r = − 87



4± 42 −4∗1∗4
(b) x + 4x = 4; so x2 + 4 = 4x and x 6= 0; so x2 − 4x + 4 = 0; so x1,2 = which gives
√ 2
us x1,2 = 2 ± 0 = 2

(c) (b − Nk )N = 0; so N = 0, or N = bk.

(d) N = 0, or b − d(1 + Nk ) = 0; so b = −d(1 + Nk ), b


d = (1 + Nk ) b
d −1 = N
k thus N = k( db − 1)
or N = k(b−d)
d .

b b
(e) N = 0, or 1+N/h − d = 0; so 1+N/h = d; so b = d(1 + N/h); so b = d + (d/h)N; b − d =
(b−d)h
(d/h)N, and thus N = d or N = h( db − 1)
4. (a) From 1st eq. x = 2y − 5, substitution into 2nd eq. gives 2(2y − 5) + y = 10 so 5y − 10 = 10,
5y = 20 so y = 4, and thus x = 2 ∗ 4 − 5 = 8 − 5 = 3.

(b) From 1st eq. x = − ba y substitution into 2nd eq. gives −cb cb
a y + dy = −b so (d − a y = −b or
da−cb −ba b b −ba b2
a y = −b so y = da−cb and hence x = − a y = − a da−bc = da−bc

(c) From 2nd eq. y(4 − x) = 0, this gives y = 0 or x = 4. Substitution of y = 0 into the 1st eq.
gives x(1 − 2x) = 0 and hence x = 0 and x = 0.5. Substitution of x = 4 into the 1st eq. of
gives 4(1 − 2 ∗ 4) + 4y = 0 so 4 ∗ −7 + 4y = 0 −28 = −4y so y = 7. Thus the solutions are:
(0, 0), (0.5, 0), (4, 7).

(d) From 2nd eq. y(9 − 3x − y) = 0, we find y = 0 or 9 − 3x − y = 0 so y = 9 − 3x. Substitution


of y = 0 into the 1st eq. gives 4x − x2 = 0 or x(4 − x) = 0 so x = 0 or x = 4. Substitution
of y = 9 − 3x into the 1st eq. gives 4x − x(9 − 3x) − x2 = 0 so 4x − 9x + 3x2 − x2 = 0 so
108 A NSWERS FOR EXERCISES

−5x + 2x2 = 0 so x(−5 + 2x) = 0 so x = 0 or x = −2.5 and thus y = 9 or y = 9 − 7.5 = 1.5


Thus the solutions are: (0, 0), (4, 0), (0, 9), (2.5, 1.5).

(e) From 2nd eq. N = 0, or R = δ. Substitution of N = 0 into the 1st eq. gives b(1 − (R/k) −
d)R = 0 so R = 0 or b(1 − (R/k) − d) = 0 so 1 − (R/k) − d = 0 1 − d = R/k so R = k(1 − d).
Substitution of R = δ into the 1st eq. gives b(1 − (δ/k) − d − aN)δ = 0 so 1 − (δ/k) −
d − aN = 0 so 1 − (δ/k) − d = aN so N = 1−d δ
a − ak Thus the solutions are: (0, 0), (k(1 −
d), 0), (δ, 1−d δ
a − ak ).

5. (a) f 0 (x) = −3x− 4 = − x34 (chain rule)

(b) f 0 (x) = −5e−5x (chain rule)

(c) f 0 (x) = (4x − x2 )0 (2x + 3) + (4x − x2 )(2x + 3)0 = (4 − 2x)(2x + 3) + (4x − x2 )(2) = 8x + 12 −
4x2 − 6x + 8x − 2x2 = 10x + 12 − 6x2 (product rule)

x0 (a2 −x2 )−x(a2 −x2 )0 x2 +a2


(d) f 0 (x) = (a2 −x2 )2
= (a2 −x2 )2
(quotient rule)

−b 2
6. (a) y-intercept so n = 0: y = − cb2 , x-intercept so y = 0 gives an n2 +c2
= 0 so an2 − b = 0 so
q
2 −b a−(b/n2 )
n = ± ba , horizontal asymptote limn→∞ an n2 +c2
= limn→∞ 1+(c2 /n2 ) = a so y = a, no division

by zero possible so no vertical asymptote.

(b) y-intercept so R = 0 gives N = hra , x-intercept so N = 0 gives R = −h and R = K, no division


by zero so no vertical asymptotes, does not approach constant number for R going to infinity
so no horizontal asymptote.

7. (a) fig.a straight line with y intercept 3 x intercept 0.5

(b) fig.b parabola with maximum, x intercepts from x(1 − 3x) = 0 so at x = 0 and x = 1/3

(c) horizontal asymptote y = 7 (for x going to infinity fraction goes to 3), y intercept, x = 0,
3x 3x
y = 4, x intercept, y = 0 x+a + 4 = 0 so x+a = −4, 3x = −4x − 4a, 7x = 4a so x = 4/7a.
(fig.c). Dependence on a. If a increases, the horizontal asymptote y = 7 will be approached
at a slower rate (i.e. you divide by larger number, so result is smaller untill large x dominates
a). (fig.c)

y
y y 7
3
0.5 x x 4
0.33
x
a b c

p
8. (a) (see below,fig.a)
p x + 3y2 = 0, 3y2 = x so y2 = (1/3)x so y = (1/3)x (upper half) and
y = − (1/3)x (lower half)
√ √
(b) (see below,fig.b) x2 + y2 = 9, y2 = 9 − x2 so y = 9 − x2 (upper half) and y = − 9 − x2
(lower half)
109

P
y y y
a
x 3 x x N

a b c d

(c) (see above, fig.c) xy = 0 two lines: x = 0 and y = 0

(d) (see above, fig.d) dN(a − P) = 0 two lines: N = 0 and P = a


NP P
(e) dN + N+a = 0 so again two lines: N = 0 and d + N+a = 0 which gives P overN + a = −d
and hence P = −d(N + a) = −dN − da which is a straight line with y intercept −da and
slope −d (see below, fig.a)
cRN cN
(f) dR(b − R) = R+a so again two lines R = 0 and d(b − R) = R+a which gives N = dc (b −
R)(R + a) which is a parabola with maximum, with x intercepts R = b and R = −a (see
below, fig.b)

P N N
N
b/a 1
−a b (d+1/k) (1+b)/e
−a
−da N R R P

a b c d

(g) bR(1 − (R/k) − dR) − aNR = 0 soR = 0 and b(1 − (R/k) − dR) − aN = 0 so N = ab (1 −
(R/k) − dR) = ab (1 − ( 1k + d)R) which is a straight line with y intercept b/a, x intercept
1
R = (1/k)+d and slope −( 1k + d) (see above, fig.c).

(h) aN + P(1 − eP) + bP = 0 so aN = −P(1 − eP) − bP = P(−1 + e − b) = P(e − (1 + b)) so


N = Pa (eP − (1 + b)) this is a parabola with minimum, with x intercepts P = 0 and P = 1+b
e
(see above, fig.d).
9. x−x0
a + y−y 0 y−y0 x−x0 x−x0 x−x0
b = 0 so b = − a , y − y0 = −b a , y = −b a + y0 which can be rewritten as
y = − ab x + ba x0 + y0 . This is a straight line with y intercept (x = 0) y = − ab x0 + y0 , x intercept
(y = 0) x = x0 + ba y0 and slope − ab .

Exercises Chapter 2
1. (a) A: more growth in y than x direction; B rotation of leaf

(b) sin(30)=0.5, cos(30)=0.87


     
0.87 −0.5 1.5 0 1.3 −1
C = B×A = × =
0.5 0.87 0 2 0.75 1.73

(c) No!
110 A NSWERS FOR EXERCISES

(d) (0,-1) → (1,-1,73);


(0,0) → (0,0);
(-1,1) → (-2.3,0.98);
(1,1) → (0.3,2.48);
(0,4) → (-4,6.92)
√ √ √
−4± 42 −4∗1∗5 −4± 16−20 −4± −4
2. (a) x1,2 = 2 = 2 = 2 = −2 ± i.

(b) x2 − 5x + 6 = 0 so (x − 3)(x − 2) = 0 so x1 = 2, x2 = 3.
    
2 −4 x 3
3. (a) = , detA = 2 ∗ 1 − (−4) ∗ 1 = 2 + 4 = 6.
1 1 y 1
    
a b x 0
(b) = , detA = ad − bc.
c d y −b
(a) (−2 − λ)(−2 − λ) 2 2
4. − 1 ∗ 1 = λ + 4λ +4 − 1 =λ + 4λ + 3 = (λ + 3)(λ + 1) = 0
−1 −1
λ1 = −1, v1 = k =k ;
−2 − (−1) −1
   
−1 −1
λ2 = −3, v2 = k =k , where k is an arbitrary number.
−2 − (−3) 1

2± 22 −4∗1∗(−3)
− λ)(1 − λ) − 4 ∗ 1 = λ2 − 2λ + 1 − 4 = λ2 − 2λ − 3 = 0 so λ1,2 =
(b) (1 √ 2 =
2± 16 2±4
2 = = 1±2
2    
−4 −4
λ1 = −1, v1 = k =k ;
1 − (−1) 2
   
−4 −4
λ2 = 3, v2 = k =k , where k is an arbitrary number.
1−3 −2

2± 22 −4∗1∗2
(c) (−1

− λ)(3 − λ) − 5 ∗ (−1) = λ2 − 3λ + λ − 3 + 5 = λ2 − 2λ + 2 = 0 λ1,2 = 2 =
2± −4 2±i2
2 = 2= 1±i
   
−5 −5
λ1 = 1 + i, v1 = k =k ;
−1 − (1 + i) −2 − i
   
−5 −5
λ2 = 1 − i, v2 = k =k , where k is an arbitrary number.
−1 − (1 − i) −2 + i

Exercises Chapter 3
dn
1. dt = 1.5 so the general solution is n(t) = Ae1.5t , from n(0) = 30 we find 30 = Ae1.5∗0 = Ae0 = A
so the solution of the initial value problem is n(t) = 30e1.5t . Next we fill in n(4) = 30e1.5∗4 =
30e60 = 12102.86. Finally we solve 60 = 30e1.5t , 2 = e1.5t , ln2 = 1.5t, giving the doubling time
t = ln2
1.5 ≈ 0.46.
dN ln2
2. dt = kN so doubling time is t = k = 20min = 20 ∗ 60 = 1200s so k = ln2/1200s ≈ 0.58 ·
10−3 sec−1
3. with N(0) = 0 equation becomes N(t) = dk (1 − e−dt ) = dk − dk e−dt . For this equation dN(t)
dt =
k −dt −dt dN(t) dN −dt k k −dt
−d ∗(− d e ) = ke now we substitute N(t) and dt into dt = kN: ke = k −d( d − d e ) =
k − (k − ke−dt ) = ke−dt and we are finished.
4. (a) Phase portrait in fig.a (below). f (x) = −15 + 8x − x2 = (−x + 3)(x − 5) is a parabola with
maximum, with x intercepts x1 = 3 and x2 = 5, attractor x = 5, basin of attraction x > 3, so
in end x = 5.
111

(b) Phase portrait also in fig.a (below). f (x) = −4 + 5x − x2 = (−x + 4)(x − 1) is a parabola
with maximum, with x intercepts x1 = 1 and x2 = 4, attractor x = 4, basin of attraction x > 1,
so in end x = 4.
(c) Phase portrait in fig.c. (below). f (x) = −x(x2 + x − 6) = −x(x + 3)(x − 2) so cubic equation
with x intercepts x = 0, x = −3, x = 2. For large x term −x3 dominates, so f (x) < 0 (allows
us to choose between 2 possible orientations of cubic graph). Thus attractor x = −3, with
basin x < 0 and attractor x = 2, basin, x > 0. So in end either x = −3 or x = 2 depending on
initial conditions.
y

x1 x2 x −3 2

a c d
3 2
(d) √
fig.d. (above).
√ √ 8x − x √= x(8 − x ) so cubic equation3 with x intercepts x = 0 x =
f (x) =
8 = 2 2 andx = − 8 = −2 2. Again, √ for large x term −x dominates, so f (x)√< 0. So
we have two attractors, attractor x =√−2 2, with√basin x < 0 and attractor x = 2 2, with
basin, x > 0. So in end either x = −2 2 or x = 2 2 depending on initial conditions.
(e) fig.a (below) attractor x = 4 basin 0 < x < 10, so in end x = 4;
fig.b (below), attractor x = 0, basin x < 1 and attractor x = 6, basin x > 1, so in end x = 0 or
x = 6 depending on initial conditions.
f(x)
f(x)

0 1 6 0 1 2
0 4 10 x
x

a b c

3x2
(f) graph is shown in fig.c (above). Function f (x) = 2+x2
− x can not be easily drawn. So find
3∗02
characteristic points to get an idea: intercept with y axis (x = 0) y = 2+02
− 0 = 0, intercept
3x2 3x2 3x 2
with x axis (y = 0) 0 = 2+x2 − x so x = 2+x2 so x = 0 or 1 = 2+x 2 , which gives us 3x = 2 + x
2
or x − 3x + 1 = (x − 2)(x − 1) = 0 and thus x = 2 or x = 1. So we have 3 intercepts with
the x axis, similar to a “normal” cubic function. But which of the two possibilities do we
have? For x going to infinity the fraction part of the function approaches 1 leaving the −x
part to go to minus infinity. So, one attractor is attractor x = 0, with basin x < 1 and the
other attractor is x = 2, with basin x > 1.
5. (a) f (x) = −x(x − 0.2)(x − 1) is a cubic function with x intercepts x = 0 x = 0.2 and x = 1. For
x large the term −x3 dominates and is below zero. The graph is shown in figure a (below).
(b) in a minimum or maximum the derivative f 0 (x) = 0, so we first rewrite f (x) = −x3 +
1.2x2 − 0.2x and find f 0 (x) = −3x2 + 2.4x − 0.2, then we find locations of √
minima/maxima
0 2 2 −12± 122 −4∗−15∗−1
from f (x) = −3x + 2.4x − 0.2 = 0 or −15x + 12x − 1 = 0 so x1,2 = 2∗−15 =
√ √ √
−12± 144−60
−30 = −12±
−30
84
= −12 84 2 9.17
−30 ± −30 = 5 ± −30 = 0.4 ± −0.31 Which of these two x values
to take? From the graph we see that at the lowest x value there is a minimum, at the highest
x value a maximum. So at the minimum x = 0.4 − 0.31 = 0.09, substituting this into f (x)
gives −0.09(0.09 − 0.2)(0.09 − 1) = −0.009
112 A NSWERS FOR EXERCISES

(c) 0.0005 < 0.009 so upward shift not enough to make first two equilibria disappear temporally,
so system will stay in domain of attraction of first equilibrium and only shift temporally with
shift of this equilibrium and then shift back with it and at the end be again at x = 0

(d) 0.02 > 0.009 so upward shift now enough to make first two equilibria disappear temporally,
so only remaining attractor is third equilibrium so the system will go there, as shift disap-
pears and first two equilibria appear again, system remains at x = 1, as moving back to x = 0
disagrees with the phase portrait.

(e) see b smax = 0.009

y y

0 0.2 1 x x
0 0.2 1

a b

6. The maximal yield occurs if h equals the maximum of parabola f (n) = rn(1 − n/k). We find it by
determining where its derivative is zero. f (n) = rn(1−n/k) = rn−(r/k)n2 so f 0 (n) = r −2(r/k)n,
from r − 2(r/k)n = 0 we find r = 2(r/k)n and hence n = k/2 substituting this in f (n) gives us
f (k/2) = r(k/2)(1 − ((k/2)/k)) = (rk/2)(1 − (1/2)) = (rk/4) which gives us hmax = rk4

Exercises Chapter 4
1. Eigenvalues from system’s matrix: (1 − λ)(8 − λ) − (−2) ∗ 5 = λ2 − 9λ + 8 + 10 = λ2 − 9λ + 18 =
(λ − 6)(λ − 3) = 0 so λ1 = 3 and λ2 = 6.      
2 2 2 2
Thus eigenvectors are v1 = = and v2 = = .
1−3 −2 1−6 −5
     
x(t) 2 2
So the general solution is = C1 e3t +C2 e6t .
y(t) −2 −5
     
3 2 3∗0 2
Now we substitute x(0) = 3 and y(0) = −3 and get = C1 e +C2 e6∗0
−3 −2 −5
     
3 2 2
which can be simplified as = C1 +C2 . We can thus write 3 = C1 ∗ 2 +
−3 −2 −5
C2 ∗ 2 and −3 = C1 ∗ −2 + C2 ∗ −5. From the first we find C1 = 1.5 − C2 . Substiting this in the
second equation gives us −3 = −2(1.5 −C2 ) − 5C2 = −3 + 2C2 − 5C2 = −3 − 3C2 so C2 = 0 and
C1 = 1.5.    
x 2
Now we can write the solution of the initial value problem as: = 1.5 e3t
y −2
2. We are dealingagainwithan initial value problem
  ofa linear system here. First write it in a more
dA
dt −0.01 0.01 A
familiar form dB = .
dt
0.04 −0.04 B
Find eigenvalues (−0.01 − λ)(−0.04 − λ) − 0.01 ∗ 0.04 = λ2 + 0.05λ + 0.0004 − 0.0004 =
λ2 + 0.05λ = λ(λ + 0.05) = 0so λ1 = 0 andλ2 =  −0.05.   
−0.01 −0.01 −0.01
Thus eigenvectors are v1 = = and v2 = =
−0.01 − 0 −0.01 −0.01 − (−0.05)
 
−0.01
.
0.04
113

     
A(t) −0.01 −0.01
So the general solution is = C1 e0t + C2 e−0.05t =
B(t) −0.01 0.04
   
−0.01 −0.01
C1 +C2 e−0.05t .
−0.01 0.04
   
3 −0.01
Now we substitute A(0) = 3 and B(0) = 5 and get = C1 +
0 −0.01
     
−0.01 −0.01 −0.01
C2 e−0.05∗0 = C1 +C2 so we can write 3 = −0.01C1 − 0.01C2
0.04 −0.01 0.04
or −300 = −C1 − C2 and 0 = −0.01C1 + 0.04C2 or 0 = −C1 + 4C2 . From the first
equation we obtain C1 = C2 − 300 substituting this into the second equation give us
0 = C2 − 300 + 4C2 = 5C2 − 300 so C2 = 60 and hence C1 = 240.   
A(t) −0.01
Now we can write the solution of the initial value problem as: = 240 +
B(t) −0.01
 
−0.01
60 e−0.05t
0.04
3. Start with simpler second equation dP d
dt = cNP − dP = 0, so P = 0 or cN − d = 0 which gives N = c .
First substitute P = 0 in first equation, which gives dN 2
dt = aN − eN = 0 so N = 0 or a − eN =
ed 2
which gives N = ae . Next substitute N = dc in first equation, which gives dN ad bd
dt = c − c2 − c P = 0
2
so bd ad ed
c P = c − c2 so bP = a − c =
ed ac−ed
c so finally P = ac−ed
bc . Thus we have a total of three
a d ac−ed
equilbria (0, 0), ( e , 0) and ( c , bc ). All non-negative if: ac ≥ ed
4. Start with simpler second equation dP c
dt = cM − dP = 0 which gives M = d P. Substitute in first
equation, which gives dM a bc a bc
dt = 1+P − d P = 0. We first reorder as 1+P = d P and then as a =
bc bc bc 2 bc 2 bc
d P(1 + P) = d P + d P , we can write this in ’abc’format as d P + d P − a = 0 and rescale as
√ √
−1± 1−4∗1∗− ad −1± 1+ 4ad
P2 + P − ad = 0. From this we find P1,2 = bc
= bc
. As proteins and mRNAs
bc 2 √ 24ad √ 4ad
−1+ 1+ bc c −1+ 1+ bc
can only have non-zero concentrations we finally get P = 2 and M = d 2

5. Again the second equation is simpler: dI α


dt = γSI − αI = 0 gives us I = 0 or S = γ . First substitute
I = 0 in the first equation, giving dS B α
dt = B − µS = 0 so S = µ . Next substitute S = γ in the first
µα µα µ
equation, giving dS α α B
dt = B − γ γ I − µ γ = B − αI − γ = 0 so αI = B − γ and thus I = α − γ . Thus
equilibria are ( Bµ , 0), and ( αγ , αB − µγ ). The first equilibrium is always positive, the second one is
positive if Bγ > αµ
6. Find M null-cline: dM a a
dt = 1+P − bM = 0 so 1+P = bM. Now note that it is much easier to write
M as a function of P than the other way around, so we simply choose to put M on the y − axis
a
and P on the x-axis. Thus we write M = b(1+P) . (NB if we would write P as a funcion of M we
would obtain P = a−bM a
bM ) How to draw the function M = b(1+P) ? Determine characteristic points.
Intercept with y axis (P = 0) occurs for M = ba . Function never becomes zero, so no x intercept.
a (a/P) a
limP→∞ b(1+P) = limP→∞ b((1/P)+1) = 0 so horizontal asymptote M = 0. limP↑−1 b(1+P) = ∞ and
a dP
limP↓−1 b(1+P) = −∞, so P = −1 is vertical asymptote. Now find the P null-cline: dt = cM − dP =
d d
0 so cM − dP = 0. Now we can write either M = or P =
cP d M.However, as we choose for the
M null-cline to write M as a function of P we have to do the same here in order to draw them in
the same graph, so we choose the first. M = dc P is a straight line going to (0, 0) and with slope
d
c . Together this gives us the shown null-clines. To determine the vectorfield, fill in either “very
big” or “very small” values for M and P. Under the P isocline M is constrained but not P, so
M is small and P is large so dP dt = cM − dP < 0, so ← so above the P isocline →. Likewise
a
above the M isocline both M and P are unconstrained, so M and P are large, so 1+P is small so
dM a
dt = 1+P − bM < 0, so ↓ and below the M iscoline ↑. Together this gives the shown vectorfield. It
is hard to tell the equilibrium type from this phase portrait, if you draw the vectorfield close to the
114 A NSWERS FOR EXERCISES

equilibrium they all point inward, suggesting stability, but the vectorfield also suggests rotation. In
the next chapter you learn that this equilibrium may be a stable spiral or a stable node.

dS B−µS
7. Find S null-cline dt = B − γSI − µS = 0 so B − µS = γSI and I = γS . To draw it determine again
B−µS
characteristic points. Intercept with S axis (I = 0) gives us γS = 0 so B − µS = 0 so S = Bµ . No
intercept with I axis, for S = 0 we get division by zero! Instead limS↑0 B−µS B−µS
γS = −∞ and limS↓0 γS =
−µ
∞ so S = 0 is a vertical asymptote. Furthermore, limS→∞ B−µS γS = limS→∞ (B§) − µoverγ = γ so
I = −µ dI
γ is a horizontal asymptote. Now find the I null-cline from dt = γSI − αI = 0, so I = 0
(horizontal line, on x axis) or S = αγ (vertical line). The figure shows the two possible phase
portraits, the left one is for Bµ < αγ and the right one is for Bµ > αγ . Now let us determine the
vectorfield. For S and I both large (upper right corner vectorfield) we can easily see that dS dt is
dominated by the −γSI term, so we need to draw ←, whereas dI dt is dominated by the γSI term,
so we need to draw ↑. From this we find the vectorfields shown below. In the left picture we see
that only the equilibrium with S non-zero and I zero is biologically relevant, and that it is a stable
node (the biologically unrealisticequilibrium with negative I is a saddle). In the right picture, we
see that the equilibrium with S non-zero and I zero is a saddle, and that the equilibrium with both
S and I non-zero shows a rotating dynamics around it. It is probably stable, as the system has no
other attractor.

Exercises Chapter 5

1−λ 4
1. (a) Characteristic eq. is det = (1 − λ)(3 − λ) − 4 ∗ 2 = λ2 − 4λ − 5 =
2 3−λ

   
−4 −4
(λ − 5)(λ + 1) = 0, so λ1 = 5 and λ2 = −1 v1 = k =k ; v2 =
−1 − 5 −4
115

   
−4 −4
k =k , thus this is a saddle point. For phase portrait see fig.a be-
1 − (−1) 2
low.

5−λ −1
(b) Characteristic eq. is det = (5 − λ)(1 − λ) − (−1) ∗ 3 = λ2 − 6λ + 8 = (λ −
3 1−λ
     
1 1 1
2)(λ − 4) = 0, so λ1 = 2, and λ2 = 4 v1 = k =k ; v2 = k =
5−2 3 5−4
 
1
k , i.e. a non-stable node. For phase portrait see fig.b below.
1

2 2± (−2)2 −4∗1∗2
(c) Similarly, here the characteristic

eq. is λ − 2λ + 2 = 0, so λ1,2 = 2∗1 =
√ √ √
2± 4−8
2 = 2± 2 −4 = 2± 42 −1 = 1 ± i so λ1 = 1 + i, λ2 = 1 − i (no eigen vectors required).
Complex eigenvalues with positive real part gives us a non-stable spiral, for the qualitative
phase portrait see fig.c below. (qualitative because we did not determine direction of rotation
from vectorfield, not that you can do that by filling in x, y values and finding the signs of dx
dt
and dy
dt !).

a b c

(d) Characteristic
  eq. is λ2 +4λ + 3 = (λ + 3)(λ + 1) = 0, so λ1 = −3, λ2 = −1 and v1 =
−1 −1
k , v2 = k i.e. a stable node. For the phase portrait see fig.a below.
1 −1

(e) Characteristic eq. is λ2 + 2λ + 2 = 0, λ1 = −1 + i, λ2 = −1 − i,(no eigen vectors required):


stable spiral. The qualitative phase portrait is given in fig.b below.

(f) Characteristic eq. is λ2 + 1 = 0, λ1 = +i, λ2 = −i,(no eigen vectors required): center point.
The qualitative phase portrait is given in fig.c below.
   
2 1 1
(g) Characteristic eq. is λ − 1 = 0, so λ1 = +1, λ2 = −1, v1 = k ; v1 = k , i.e.
−3 −1
a saddle . The phase portrait is given in fig.d below.

a b c d

2. (a) dx
dt = x(1 − x − y) = 0 gives us x = 0 and y = 1 − x as x null cline, dy dt = y(1 − 2x) gives
us y = 0 and x = 0.5 as y null cline (see Fig.). Equilibria are (from intersection points)
(0, 0) (1, 0) and (0.5, 0.5). Fill in point (1, 1) gives dx
dt = 1(1 − 1 − 1) = −1 < 0 so ← and
dy
dt = 1(1 − 2 ∗ 1) = −1 < 0 so ↓. From this follows rest of vectorfield. Now we can see
that (0, 0) and (0.5, 0.5) are saddle points and (1, 0) is stable node. Thus the system goes to
(1, 0).
116 A NSWERS FOR EXERCISES

dN
(b) dt = 2N − NP − N 2 = N(2 − P − N) = 0 gives us N = 0 and P = 2 − N as N null cline.
dP
dt = 3P − 2NP − P2 = P(3 − 2N − P) = 0 gives us P = 0 and P = 3 − 2N as P null cline
(see Fig). Equilibria are (from intersection points) (0, 0), (2, 0), (0, 3) and (1, 1). Fill in point
(2, 2) gives us dN dP
dt = 2(2 − 2 − 2) = −4 < 0 so ← and dt = 2(3 − 2 ∗ 2 − 2) = −6 < 0 so ↓.
From this follows rest of vectorfield. Now we can see that (0, 0) is an unstable node, (1, 1)
is a saddle node which separatrix forms the boundary between the domains of attraction of
the stable nodes (2, 0) and (0, 3). Depending on the initial conditions, in which domain of
attraction these ly, the system thus goes to either (2, 0) or (0, 3). Note that this is a model
for competetive exclusion, in which either one or the other species can be present.

3. The system is linear, so there is only one equilibrium at (0, 0) You can also find this from
the equations: from the second 3x − y = 0 so y = 3x fill in in the first equation −2x −
a ∗ 3x = 0 = −(2 + 3a)x = 0 so x = 0 and hence y = 0. The characteristic equation is
−2 − λ −a
given by det = (−2 − λ)(−1 − λ) − (−3)a = λ2 + 3λ + 2 + 3a = 0, λ1,2 =
3 −1 − λ
√ √ √
−3± (−3)2 −4∗1∗(2+3a) −3± 9−8−12a −3± 1−12a
2 = 2 = 2 .
1
If 1 − 12a < 0, i.e. a > 12 we have complex roots. Because the real part for complex roots

is
1 −3− 1−12a
negative (−3/2) we will have a stable spiral.√If the roots are real (a >= 12 ), √
then λ2 = 2
is always negative. The first root λ1 = −3+ 21−12a will be positive if −3 + 1 − 12a > 0, i.e. if
the root is larger than 3 and hence if the expression under the root is larger than 9 , this gives us
1 − 12a > 9 so −12a > 8 so 12a < −8 or a < −8 2 2
12 = − 3 . Thus if a < − 3 , then λ1 > 0 and λ2 < 0
and we have a saddle point. In the interval − 32 < a < 12
1
we have λ1 < 0 and λ2 < 0 so we have a
stable node.
Conclusion, if we increase a from −∞, we will first have a saddle point (unstable eq.) till a = − 23 ,
1 1
then the saddle will become a stable node (stable eq.), until a = 12 . If a becomes more than 12 we
will have a stable spiral (stable eq.).

−a − λ 1
4. The characteristic equation of this system is given by det = (−a − λ)(−b −
(2a − 3) −b − λ

−(a+b)± (a+b)2 −4∗1∗(ab+3−2a)
λ)−1∗(2a−3) = λ2 +(a+b)λ+ab+3−2a = 0. So we get λ1,2 = 2 .
117

We only need to know when a center point occurs, which is when the root is complex and has a
zero real part. This means that (a + b)2 − 4 ∗ 1 ∗ (ab + 3 − 2a) < 0 and a + b = 0 so b = −a. Filling
in b = −a in (a + b)2 − 4 ∗ 1 ∗ (ab + 3 − 2a) < 0 gives 4a2 + 8a − 12 < 0 or a2 + 2a − 3 < 0 or
(a + 3)(a − 1) < 0 For a = −3 or a = 1 this equation is exactly zero. Now we need to figure out if
the equation is < 0 for a between −3 and 1 or for a < −3 or a > 1. For this, fill in a value eg. a = 2
(2 + 3)(2 − 1) = 5 ∗ 1 = 5 > 0. So indeed for (a + 3)(a − 1) < 0 we need to have −3 < a < 1. So
the conditions we have for a center are: b = −a and −3 < a < 1.
 dx
5. (a) System is dt = −(a + c)x + by
dy
dt = ax − (b + e)y
 dx
dt = −5x + 2y −5 − λ 2
(b) System now is dy Characteristic equation is det =
dt = 0.5x − 5y
0.5 −5 − λ
(−5 − λ)(−5 − λ) − 2 ∗ 0.5 = λ2 + 10λ + 25 − 1 = λ2 + 10λ + 24 = (λ + 4)(lambda + 6) = 0
so eigen values λ1 = −4, λ2 = −6, so stable node.

−(a + c) − λ b
(c) Characteristic equation in general case is det = (−(a + c) −
a −(b + e) − λ
λ)(−(b + e) − λ) − b ∗ a = λ2 + (a + b + c + e)λ 2
√+ ab + ae + cb + ce − ab = λ + (a + b + c +
−(a+b+c+e)± (a+b+c+e)2 −4∗1∗(ae+cb+ce)
e)λ + ae + cb + ce = 0 so λ1,2 = 2 . As all parameters
are > 0 we know that a + b + c + e > 0 and hence that −(a + b + c + e) < 0. However
depending on the values of the parameters the value under the root sign can be either positive
or negative.
So if (a + b + c + e)2 > 4(ae + cb + ce), we have a real valued solution and the value of the
root is (a + b + c + e)2 − 4(ae + cb + ce) so < (a + b + c + e) so λ1 < 0 and λ2 < 0 so we
have a stable node. If (a + b + c + e)2 < 4(ae + cb + ce), the root is complex, the real value
of the complex eigenvalues is −(a + b + c + e) < 0 so we have a stable spiral.

Exercises Chapter 6
∂z ∂z
1. (a) ∂x = 2x at (1, 2) it is 2; ∂y = 2y at (1, 2) it is 4;

(b) ∂z
∂x = 25 − 3x2 − y2 , at (3, 4) it is -18;

(c) ∂z
∂N = bR − d, at 0, 0 it is −d; at R = db , N = 1 it is 0; ∂z
∂R = bN, at 0, 0 it is 0; at R = db , N = 1
it is b.
∂z a ∂z
(d) ∂P = − (1+P)2, ∂M = −b;
∂z ∂z
(e) ∂N = a − 2eN − bP, ∂P = −bN;
∂z νA ∂z νMh
(f) ∂M = L − h+A , ∂A = −δ − (h+A)2;

∂z 2aP1 P2 ∂z a(h+P2 )
(g) ∂P1 = (h+P12 +2P2 )2
and ∂P2 = − (h+P2 +2P
1
)2
1 2

∂z b2 NT (2+cN) ∂z b2 N 2 (1+cN()
(h) ∂N = (1+cN+bT N 2 )2 ∂T
= (1+cN+bT N 2 )2
∂f ∂f
2. partiaxx= 2x so in (1, 1) this is 2, partiaxy = 2y so in (1, 1) this is 2 and f (1, 1) = 2 so f (x, y) ≈
2 + 2(x − 1) + 2(y − 1) = 2 + 2x − 2 + 2y − 2 and hence f (x, y) ≈ −2 + 2x + 2y
3. (a) dx
dt = −4y = 0 gives y = 0 substituting this into dy 2
dt = gives 4x − x = x(4 − x) = 0 so x = 0
or x = 4. Thus equilibria are (0, 0), (4, 0). ∂∂xf = 0, ∂∂yf = −4, ∂g ∂g
∂x = 4 − 2x, ∂y = −0.5 so in
(0, 0):0, −4, 4, −0.5; and in (4, 0): 0, −4, −4, −0.5.
118 A NSWERS FOR EXERCISES

dy
(b) dt = x − y = 0 gives x = y substituting this into dx 2
dt = 0 gives us 9x + x = x(9 + x) = 0 so
x = 0 or x = −9. Thus equilibria are (0, 0) and (−9, −9). ∂∂xf = 9 ∂∂yf = 2y ∂g ∂g
∂x = 1 ∂y = −1
so in (0, 0): 9, 0, 1, −1; and in (−9, −9): 9, −18, 1, −1.

(c) dx
dt = 2x − xy = x(2 − y) = 0 gives us x = 0 or y = 2. Substituting x = 0 into dy
dt = 0 gives
dy
us −y = 0 so y = 0. Substituting y = 2 into dt = 0 gives us −2 + 4x = 0 so x = 0.5. So
equilibria are (0, 0) and (0.5, 2) ∂∂xf = 2 − y ∂∂yf = −x ∂g 2 ∂g
∂x = y ∂y = −1 + 2xy so in (0, 0):
2, 0, 0, −1 and in (0.5, 2): 0, −0.5, 4, 1

(d) dx
dt = (1 − x − 3y)x = 0 so x = 0 or x = 1 − 3y. Substituting x = 0 into dy dt = 0 gives
(1 − 2y)y = 0 so y = 0 or y = 0.5. So now we already have equilibria (0, 0) and (0, 0.5).
Now we substitute x = 1 − 3y into dy dt = 0 which gives (1 − 2(1 − 3y) − 2y)y = 0 so
(1 − 2 + 6y − 2y)y = 0 or (−1 + 4y)y = 0 so y = 0 and hence x = 1 or y = 0.25 and
hence x = 0.25. This gives us equilibria (1, 0) and (0.25, 0.25). So we now have four
equilibria: (0, 0), (0, 0.5), (1, 0), (0.25, 0.25). f (x, y) = (1 − x − 3y)x = x − x2 − 3xy g(x, y) =
(1 − 2x − 2y)y = y − 2xy − 2y2 so ∂∂xf = 1 − 2x − 3y ∂∂yf = −3x ∂g ∂g
∂x = −2y ∂y = 1 − 2x − 4y. So
in (0, 0):1, 0, 0, 1; in (0, 0.5): −0.5, 0, −1, −1; in (1, 0):−1, −3, 0, −1; and in (0.25, 0.25):
−0.25, −0.75, −0.5, −0.5.
4. (a) f (x, y) = x + 4y + ex − 1 so f (0, 0) = 0 + 4 ∗ 0 + e0 − 1 = 0 + 0 + 1 − 1 = 0 g(x, y) = −y − yex
so g(0, 0) = −0 − 0 ∗ e0 = 0 − 0 ∗ 1 = 0
∂f ∂f
(b) ∂x = 1 + ex ∂y =4 ∂g
∂x = −yex ∂g
∂y = −1 − ex
∂f 0 ∂f ∂g 0 ∂g 0
(c) ∂x (0, 0) = 1 + e = 1+1 = 2 ∂y (0, 0) = 4 ∂x (0, 0) = −0e = 0∗1 = 0 ∂y (0, 0) = −1 − e =
−1 − 1 = −2
dx dy
(d) dt = 0 + 2(x − 0) + 4 ∗ (y − 0) = 2x + 4y and dt = 0 + 0 ∗ (x − 0) − 2 ∗ (y − 0) = −2y
5. Note that these systems are already linear. Thus, they have a single equilibrium point at (0, 0) and
we already have them in matrix form and can directly start with finding eigenvalues and eigenvec-
tors.
(a) Characteristic equation is (−2 − λ)(−2 − λ) − 1 ∗ 1 = λ2 + 4λ + 3 = (λ  + 1)(λ + 3) =
 0.
−1
So eigenvalues are λ1 = −1 and λ2 = −3 and eigenvectors are v1 = k =
−2 − (−1)
       
−1 −1 −1 x(t)
k v2 = k =k . So the general solution is =
−1 −2 − (−3) 1 y(t)
   
−1 −1
C1 e−1t +C2 e−3t and the equilibrium is a stable node.
−1 1

(b) Characteristic equation is (3 − λ)(4 − λ) − (−1) ∗ (−2) = λ2 − 7λ + 12 − 2 = λ2 − 7λ +


10 = (λ− 5)(λ −2) = 0. So eigenvalues
  1 =5 and
are lambda  λ2 = 2 and eigenvectors are
1 1 1 1
v1 = k =k v2 = k =k . So the general solution is
3−5 −2 3−2 1
     
x(t) 1 1
= C1 e5t +C2 e2t and the equilibrium is an unstable node.
y(t) −2 1

Exercises Chapter 7
1. (a) detA = 3 ∗ 6 − 1 ∗ −20 = 18 + 20 = 38 > 0 so no saddle, trA = 3 + 6 = 9 > 0 so unstable,
D = trA2 − 4detA = 92 − 4 ∗ 38 = 81 − 152 = −71 < 0 so complex solutions so unstable
spiral.
119

(b) detA = 2 ∗ −10 − 1 ∗ 2 = −20 − 2 = −22 < 0 so saddle so unstable.


(c) detA = 2 ∗ −2 − 1 ∗ 5 = −4 − 5 = −9 < 0 so saddle so unstable.
(d) detA = 1 ∗ −1 − 10 ∗ −10 = −1 + 100 = 99 > 0 so no saddle, trA = 1 − 1 = 0 so neutrally
stable center point.
2. (a) dx
dt = 2x(1 − y) = 0 gives x = 0 or y = 1. First substitute x = 0 in dy dy
dt to get dt = 2 − y = 0
dy dy
which gives y = 2. Next substitute y = 1 in dt to get dt = 2 −1 −x2 = 1 −x2 = 0 which gives
x2 = 1 and hence x = 1 or x = −1. So equilibria are (0, 2), and (1, 1). Note that equilibrium
(−1, 1) is not valid as algea can not be < 0.
y) = 2 − y − x2 then∂ f /∂x = 2(1 − y), ∂ f /∂y = −2x,
(b) If we take f (x, y) = 2x(1 − y) and g(x, 
2(1 − y) −2x
∂g/∂x = −2x and ∂g/∂y = −1 so J =
−2x −1
 
−2 0
(c) For (0, 2) we get J = so detJ = −2 ∗ −1 = 2 > 0, trJ = −2 − 1 = −3 < 0
0 −1
so stable, D = trJ 2 − 4detJ = 9 −4 ∗ 2 = 9 − 8 = 1 > 0 so real numbers so stable node.
0 −2
For (1, 1) we get J = so detJ = (0 ∗ −1) − (−2 ∗ −2) = −4 < 0 so saddle:
−2 −1
unstable.

(d)
3. (a) dx
dt = 2x(1 − y) = 0 gives x = 0 or y = 1 as x null clines. dy 2
dt = 2 − y − x = 0 gives y = 2 − x
2

as y null cline. With no free parameters we can simply fill in numbers here. Let us fill in
dy
x = 2 and y = 2. This gives dx 2
dt = 2 ∗ 2(1 − 2) = 4 ∗ −1 = −4 so ← and dt = 2 − 2 − 2 = −4
so ↓. From this we find the rest of the vectorfield.

(b)
 
−α 0
(c) For (0, 2) we get J = (note that zero in second row occurs because if you
0 −γ
move to the left only for a tiny bit the parabola is horizontal and you end up exactly on the
120 A NSWERS FOR EXERCISES

y null cline) so detJ = −α ∗ −γ − 0 ∗ 0 = α ∗ β > 0, trJ = −α − γ = −(α +γ) < 0 so stable,



0 −β
but not possible to tell whether it is spiral or node. For (1, 1) we get J = so
−δ −γ
detJ = (0 ∗ −γ) − (−β ∗ −δ) = −β ∗ γ < 0 so saddle: unstable.

(d) As stated, we find the same answers, except that for the stable equilibrium we can not deter-
mine whether it is node or spiral.

y y y 2
1 2

2x 3 1 3
1 x x

a b c
1 2
2
1 3
d e f

4. (a) graphical Jacobian: at 1 stable node. At 2 saddle. Null-clines fig.a above, phase portrait
fig.d.

(b) At 1 saddle. At 2 not known, stable node/spiral. At 3 saddle. Null-clines fig.b above, phase
portrait fig.e.

(c) At 1 not known, stable node/spiral. At 2 saddle. At 3 not known, stable node/spiral. Null-
clines fig.c above, phase portrait fig.f.
5. (a) From dNdt = aN − bNP = N(a − bP) = 0 we find N = 0 and P = a/b. Substituting N = 0
into dP dP dP
dt = 0 gives us dt = −dP = 0 so P = 0. Substituting P = a/b into dt = 0 gives us
dP
dt = cN(a/b) − d(a/b) = 0 and hence cN − d = 0 and hence N = d/c. Thus the non-trivial
equilibrium is ((d/c), (a/b)).

(b) From f (N, P) = aN − bNP we find ∂ f /∂N = a − bP, ∂ f /∂P = −bN, while from
 P) = cNP − dP 
g(N, we find ∂g/∂N = cP, ∂g/∂P = cN − d so J in ((d/c), (a/b)) J =
0 −(bd)/c
.
(ac)/b 0
(c) detJ = 0 ∗ 0 − (−b ∗ (ac)/b) = ac > 0 so no saddle. trJ = 0 so center point, neutrally stable.

(d) N and P oscillate over time, the amplitude of the oscillations depend on the initial conditions.
121

6. (a) See Figure

 
+α 0
(b) (0, 0),graphical Jacobian: J = so detJ = −α ∗ γ < 0 so saddle; ( dc , ab ), graph-
0 −γ
 
0 −β
ical Jacobian: J = so detJ = −(−β ∗ δ) = β ∗ δ > 0 and trJ = 0 so center.
+δ 0
See Figure for phase portraits

(c) In this case same results from analytical and graphical approach

(d) No, for positive parameter values (0, 0) remains a saddle and ( dc , ab ) remains a center point
independent of how large a, b, c, d exactly are.
7. (a) Null-clines
 are given  below in fig.a, thus the graphical Jacobian (on basis of ’black’ points)
α β
is: J = . This gives for detJ = −αδ + βγ, so we can not determine the
−γ −δ
sign. Thus graphical
 Jacobian does not work here. The analytical Jacobian here is:
3 1
J= , that gives detJ = −2 < 0, thus we have a saddle point. Phase portrait is
−1 −1
in fig.b.

a b c d

 
α β
(b) Null-clines are given above in fig.c, Graphical Jacobian is J = so detJ = −α∗
−γ −δ
 −1 
+β ∗ δ and trJ = α − γ and we can not find signs. Analytical Jacobian is J =
−2 −2
so detJ = (1 ∗ −2) − (2 ∗ −2) = −2 + 4 = 2 > 0 and trJ = 1 + −2 = −1 < 0 so stable spiral.
Phase portrait is given in fig.d.
 
0 β
(c) Null-clines are given below in fig.a, the graphical Jacobian gives J = so detJ =
γ 0
−β ∗ γ < 0 so saddle point. Phase portrait shown in fig.b.
 
α −β
(d) Null-clines are given below in fig.c, the graphical Jacobian gives J = so detJ =
γ δ
α ∗ δ − (−β ∗ γ) = α ∗ δ + β ∗ γ > 0 so no saddle. trJ = α + δ > 0 so unstable, but
 we can not

1 −4
determine whether it is a node or a spiral. The analytical Jacobian is gives J =
1 1
122 A NSWERS FOR EXERCISES

so detJ = 1 ∗ 1 − (−4 ∗ 1) = 1 + 4 = 5, trJ = 1 + 1 = 2 and D = trJ 2 − 4detJ = 22 − 4 ∗ 5 =


4 − 20 = −16 < 0 so complex numbers so unstable spiral. Phase portrait shown in fig.d.

a b c d

Exercises Chapter 8
1. (a) A stable limit cycle to separate the trajectories coming from the saddle and those coming
from the unstable spiral. See figure below.
(b) Point A lies inside the stable limitcycle, where the unstable spiral is. Thus, the trajectory
will spiral outward until the stable limitcycle is reached, then it will stay on the limitcycle.
Thus x will show oscillatory dynamics that first increase in amplitude and then stay constant.
See figure below.

y −
3−
− A
1−
| | | | | | |
1 3 5 x

6 x

1 t

2. In fig.8.8b the inner limitcycle must be unstable, and inside it must be a stable equilibrium, prob-
ably a spiral. Thus point A spirals inward to this stable equilibrium, showing oscillations with
decreasing amplitude untill a constant non-oscillating level is reached. Point B spirals inward to
the stable outer limitcycle, thus also showing oscillations with decreasing amplitude, but now till a
constant amplitude of oscillations is reached. Point C spirals outward to the stable outer limitcycle,
thus shoing oscillations with increasing amplitude untill the limit cycle is reached and oscillations
attain a constant amplitude. See two left figures below.
In fig.8.8c the inner limit cycle must be stable, and the equilibrium inside it must be unstable, prob-
ably a spiral. So now point A spirals outward to the inner stable limitcycle, oscillation amplitudes
first increase and then become constant. Point C spirals inward to the inner stable limitcycle, with
oscillation amplitude first decreasing and then becoming constant. Point B moves away from the
unstable outer limitcycle. See two right figures below.
3. We know that λ1,2 = α ± iβ with α = 0. We also know that trJ = λ1 + λ2 and detJ = λ1 × λ2 . Thus
we can write trJ = iβ − iβ = 0 and detJ = iβ × −iβ = −i2 β2 = −(−1)β2 = β2 > 0.
4. (a) Start with the simpler second equation: dy 2
dt = bx − x y = 0 gives x = 0 or b − xy = 0 so y =
b/x. First fill in x = 0 in the first equation dx
dt = a 6= 0 so this can never give an equilibrium!
Now fill in y = b/x in the first equation dx 2
dt = a − (b + 1)x + x (b/x) = a − bx − x + bx =
a − x = 0 so x = a. So the non-trivial equilibrium is a, (b/a).
123

y − B
y −B 3−

C A
3−

C A 1−
1− | | | | | |
1 4 7 x
| | | | | | x B
1 3 5 x
8
6
C
C
A A
B
1 1 t

(b) To do this we need to find the Jacobian in the equilibrium point. f (x, y) = a − (b + 1)x + x2 y
so ∂ f /∂x = −(b+1)+2xy = −(b+1)+2a(b/a) = −(b+1)+2b = b−1 and ∂ f /∂y = x2 =
a2 . g(x, y) = bx − x2 y so ∂g/∂x = b − 2xy = b − 2a(b/a)
 = b − 2b =2 −b  and∂g/∂y = −x 
2=

b−1 a b−1 1
−a2 . Thus the Jacobian in the equilibrium is: J = = .
−b −a2 −b −1
So trJ = b − 2 detJ = (b − 1) ∗ −1 − 1 ∗ (−b) = −b + 1 + b = 1. Thus, the determinant is
positive so the equilibrium can never be a saddle point. Stability thus fully depends on trJ.
For b > 2 trJ > 0 and the equilibrium is unstable, for b < 2 trJ < 0 and the equilibrium is
stable and for b = 2 trJ = 0 and the equilibrium is a neutrally stable center point.

(c) trJ = 0 for b = 2.

(d) Stable constant amplitude oscillations require the presence of a stable limitcycle. A stable
limitcycle implies that the equilibrium inside it must be unstable so trJ > 0 so b > 0.
5. (a) P null-cline: dP
dt = rP(1−P/K)−aRP/(h+P) = 0 so P = 0 or r(1−P/K)−aP/(h+P) = 0
which can be rewritten as r(1 − P/K) = aP/(h + P) or r(1 − P/K)(h + P) = aP and thus
finally as P = (r/a)(1 − P/K)(h + P) which is parabola with intercepts with the x axis at
P = K and P = −h and a maximum at P = (K − h)/2. R null-cline: dR dt = caRP/(h + P) −
dR = 0 so R = 0 or caP/(h + P) − d = 0 which can be written as caP/(h + P) = d, so
caP = d(h + P) = dh + dP and thus (ca − d)P = dh and finally P = dh/(ca − d), which is
simply a vertical line. So the question is whether this R null-cline intersects the P null-cline
parabola left or right of the top: i.e whether dh/(ca − d) is smaller or bigger than (K − h)/2.
For the given parameter values dh/(ca − d) = 2.5 ∗ 0.1/(1 ∗ 3 − 2.5) = 0.25/0.5 = 0.5. For
K = 0.8 we get (K −h)/2 = (0.8−0.1)/2 = 0.7/2 = 0.35 so intersection right of the top (left
figure below). For K = 1.6 we get (K − h)/2 = (1.6 − 0.1)/2 = 1.5/2 = 0.75 so intersection
left of the top (right figure below).

 
−α −β
(b) For the left figure we get the graphical Jacobian: J = so trJ = −al pha and
+δ 0
detJ = βδ. Thus the equilibrium is stable.
124 A NSWERS FOR EXERCISES

 
+α −β
For the right figure we get the graphical Jacobian: J = so trJ = −al pha and
+δ 0
detJ = βδ. Thus the equilibrium is unstable.

(c) Oscillations require a stable limitcycle, which require an unstable equilibrium inside. Thus
oscillations require intersection left of the top, so dh/(ca − d) < (K − h)/2 so 0.5 < 0.5K −
0.05 so 1 < K − 0.1 so 1.1 < K so K > 1.1

(d) See before: dh/(ca − d) < (K − h)/2 so 2dh/(ca − d) < K − h so K > 2dh/(ca − d) + h.