You are on page 1of 64

MATH259 Mathematical Modelling

Yuri Litvinenko
June 24, 2011
Contents
1 First-order ordinary dierential equations 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Exponential growth and decay . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Separable and linear equations . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Equilibrium and stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Second-order ordinary dierential equations 19
2.1 Linear second-order equations . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 The homogeneous equation with constant coecients . . . . . . . . . . . . 21
2.3 Linear oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Equilibrium and stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Inhomogeneous equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Elements of classical mechanics 33
3.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Dynamics and gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Work and energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Planetary orbits and Keplers laws . . . . . . . . . . . . . . . . . . . . . . 40
4 Systems of rst-order dierential equations 47
4.1 Homogeneous linear systems with constant coecients . . . . . . . . . . . 47
4.2 Equilibrium and stability: linear systems . . . . . . . . . . . . . . . . . . . 49
4.3 Equilibrium and stability: nonlinear systems . . . . . . . . . . . . . . . . . 50
5 Dierence equations 57
5.1 Geometric growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 The equation x
k+1
= ax
k
+ b . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.3 Fibonacci numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Nonlinear dierence equations . . . . . . . . . . . . . . . . . . . . . . . . . 60
1
Chapter 1
First-order ordinary dierential
equations
1.1 Introduction
Mathematical modelling is a part of applied mathematics. We use mathematical mod-
elling to describe and predict phenomena in the world around us. We may be interested
in the temporal evolution of a phenomenon, its steady state, or its stability. Mathemat-
ical modelling always involves both mathematics and some area of science. To isolate
mathematics from the practical demands of the sciences is to invite the sterility of a cow
shut away from the bulls (P. L. Chebyshev).
A dierential equation is an equation involving an unknown function and its derivatives
with respect to one or more independent variables. An ordinary dierential equation
contains only one independent variable. For example, a rst-order ordinary dierential
equation for a function x(t)
dx
dt
= f(t, x)
contains its rst derivative or the rate of change
dx
dt
= lim
t0
x(t + t) x(t)
t
.
The notation x = dx/dt is used when the independent variable t is time.
Dierential equations is the basic tool for the study of change in the physical world.
In this course we focus on the use of ordinary dierential equations in modelling.
Example. The speed of an object moving along the x-axis is given by v = x. If we know
v(t) and initially x(0) = 0, we can determine the position of the object at a later time t
by integration:
x(t) =
_
t
0
dx
ds
ds =
_
t
0
v(s)ds,
where s is a dummy variable of integration. Sometimes the denite integral is also written
less formally as
_
t
0
v(t)dt.
2
A model consists of a dierential equation and an initial condition, say x(0) = x
0
. We
expect that the general solution of a rst-order equation contains an arbitrary constant
of integration. There are exceptions: for instance x = 0 is the only real solution to
x
2
+ x
2
= 0. Such equations, however, are very unlikely to be encountered in practice.
Therefore we typically use an initial condition to specify the integration constant and thus
obtain a unique solution.
Example. Paradichlorobenzene (moth repellent) sublimes from the solid to the gaseous
state. We want to predict the radius r(t) of a mothball at any time if initially r(0) = r
0
.
Observationally, the volume V = 4r
3
/3 of a spherical mothball decreases so that dV/dt
is proportional to the surface area A = 4r
2
of the ball. Thus observations imply that

dV
dt
= kA,
where k > 0 is a proportionality constant. We have
4r
2
r = k4r
2
or, as long as r > 0,
r = k,
which we integrate to obtain
r(t) = C kt = r
0
kt,
where C is an integration constant that is specied by the initial condition r(0) = r
0
= C.
Finally, r(t
s
) = 0 denes the sublimation time t
s
= r
0
/k. Because the radius r(t) cannot
be negative, the correct solution for t > t
s
is simply r(t) = 0.
1.2 Exponential growth and decay
Suppose the rate of change of x(t) is proportional to x itself:
x = ax, a = const.
We obtain the general solution to this equation by rewriting it in the form of dierentials
and integrating:
dx
x
= adt,
ln x = ln C + at, C = x(0),
or
x(t) = x(0)e
at
,
where the integration constant C is specied by an initial condition. More generally,
x(t) = x(t
0
)e
a(tt
0
)
,
3
where x(t
0
) is the value of x at some time t
0
.
Solutions with a > 0 describe unlimited growth, x as t . T. R. Malthus
(1798) emphasised the role of exponentially growing solutions in models for population
growth, so such solutions are sometime said to describe the Malthusian growth. In reality,
exponential growth is limited by nite resources. Still, exponentially growing solutions
help us understand how quickly things get out of hand. Solutions with a < 0,
x(t) = x(0)e
|a|t
,
describe exponential decay, x 0 as t .
It is often useful to dene a doubling time t
2
such that x(t
2
)/x(0) = 2. For an
exponentially growing x(t), exp(at
2
) = 2 and we obtain
t
2
=
ln 2
a

0.7
a
, x(t) = x(0)2
t/t
2
.
Example. Presently the population of New Zealand grows annually by about 1%. If the
trend continues, the population will double in about 0.7/0.01 = 70 years.
Example. To double an investment in 8 years, it is necessary to achieve an annual return
exceeding 0.7/8 0.09 = 9%.
Similarly, we dene a half-life time t
1/2
for a decreasing function, say the amount of a
radioactive material, x(t
1/2
)/x(0) = 1/2. For an exponentially decaying x(t),
t
1/2
=
ln 2
|a|

0.7
|a|
, x(t) = x(0)2
t/t
1/2
.
Example. Radium decays with a half-life time of about 1500 years, so after 1000 years
there remains 2
1000/1500
= 2
2/3
0.63 = 63% of the initial amount of radium.
Example. Carbon-14 dating, discovered by W. Libby around 1949, is one of the most
accurate ways of dating archaeological nds. Radioactive carbon-14 (
14
C) is continuously
produced in the atmosphere of the earth by cosmic rays. This radiocarbon is incorporated
in carbon dioxide and absorbed by plants. Animals, in turn, build radiocarbon by eating
the plants. In living tissue, the rate of ingestion of
14
C balances the rate of disintegration
of
14
C. When an organism dies (t = 0), its
14
C concentration starts to decrease:
N(t) = N(0)e
t
,
where is the decay constant of
14
C, corresponding to t
1/2
5700 years. Consequently
the rate of decay is given by
R(t) =

N = N(0)e
t
= R(0)e
t
.
4
We can measure the present rate R(t) of disintegration of
14
C in a sample, say charcoal
from an archaeological site. Because the initial decay rate R(0) is equal to the rate of
decay of
14
C in the same amount of living tissue, we can determine the age t of the sample:
t =
1

ln
_
R(0)
R(t)
_
= t
1/2
ln[R(0)/R(t)]
ln 2
.
Consider next a generalisation of the equation for exponential growth or decay:
x = ax + b, a = const, b = const.
This is an example of a linear (meaning that the equation does not contain terms nonlinear
in x, like x
2
, sin x or x x), rst-order dierential equation with constant coecients. If b =
0, the equation is called inhomogeneous (or nonhomogeneous). If b = 0, it is homogeneous.
We note that x = b/a is a particular solution of this equation, and we seek its general
solution in the form x(t) = y(t) b/a. Putting this into the dierential equation yields
y = ay
with the general solution
y(t) = Ce
at
, C = y(0) = x(0) +
b
a
.
Hence
x(t) =
_
x(0) +
b
a
_
e
at

b
a
.
Linear rst-order ordinary dierential equations with constant coecients are often
encountered in practice. A few examples follow.
Example. Consider an RC electric circuit in which the resistance is R and the capaci-
tance is C (not to be confused with an integration constant). The governing equation for
the electric current I(t) in the circuit is
R

I +
I
C
= 0, I(0) = I
0
,
so
I(t) = I
0
exp
_

t
RC
_
.
Example. Consider an electric circuit consisting of an inductor of inductance L, a resistor
of resistance R, and a battery producing a constant electromotive force (voltage) E. The
governing equation for the electric current I(t) in the circuit is
L

I + RI = E.
5
This is a linear rst-order equation with the general solution
I(t) =
E
R
+ C exp
_

Rt
L
_
.
If I(0) = 0, we obtain
I(t) =
E
R
_
1 exp
_

Rt
L
__
.
We see that I E/R as t . Hence Ohms law E = RI is nearly true for large t.
Example. The rate of cooling of some body in air is approximately proportional to the
dierence between the temperature of the body and the temperature of the air. Suppose
we move a thermometer from a warm room (temperature T
0
) to a cooler room (temper-
ature T
1
< T
0
). The thermometer reading T(t) is governed by

T = k(T T
1
), T(0) = T
0
,
where k is a proportionality constant. The general solution is
T(t) = T
1
+ Ce
kt
.
We use the initial condition T
0
= T
1
+ C to nd the integration constant C. Hence
T(t) = T
1
+ (T
0
T
1
)e
kt
.
It makes sense that T approaches T
1
(the temperature in the cooler room) for large t.
Example. In some chemical reactions, the rate of conversion of a substance is propor-
tional to the amount of available substance. This is the case, for instance, when calcium
carbonate releases carbon dioxide on heating above 840

C, to form calcium oxide, com-


monly called quicklime:
CaCO
3
CaO + CO
2
.
If x(t) is the amount of the substance that has been converted in the time interval t, and
a is the initial quantity of the substance, then we have
x = k(a x), x(0) = 0,
where k is a constant. Integrating this equation, we obtain
x(t) = a(1 e
kt
).
Consequently, x a as t .
Example. Suppose you borrow N
0
dollars at an interest rate to be paid o at a constant
rate Q over time T. What is Q? For continuously compounded interest, the amount N(t)
that you owe at time t is governed by the equation

N = N Q,
6
and its general solution is
N(t) =
Q

+ Ce
t
.
Now two conditions are needed to specify both Q and an integration constant C. We
require N(0) = N
0
and N(T) = 0. Therefore,
N(0) = N
0
=
Q

+ C, N(T) = 0 =
Q

+ Ce
T
.
On solving these two equations, we nd C = (Q/) exp(T) and
Q =
N
0
1 e
T
.
It follows that
N(t) =
Q

_
1 e
(tT)
_
= N
0
e
T
e
t
e
T
1
.
The total amount you will pay, QT, is typically signicantly greater than the borrowed
amount N
0
. For instance, if = 6% per year and T = 25 years,
QT
N
0
=
T
1 e
T
1.9.
We can also calculate the fraction of the rst payment which is interest:
N
0
dt
Qdt
=
N
0
Q
= 1 e
T
.
Not surprisingly, the fraction is close to 1 for most loans.
Example. A patrol craft sights a surfaced enemy submarine that immediately submerges
and proceeds on an unknown straight course. The craft is twice as fast as the sub. What
path should the patrol craft follow to assure that it passes directly over the sub? To
answer this question, it is convenient to use polar coordinates r and . Choosing the
origin at the initial location of the sub, we have r for its speed. Because the craft is twice
as fast as the sub, we have
ds
dr
=
s
r
= 2,
where s is the arc length of the curve, followed by the craft, and s is the speed of the
craft. We choose the radial speed r to be the same for both the craft and the sub. Now,
if we substitute x = r cos , y = r sin into ds =
_
(dx)
2
+ (dy)
2
, we get
ds =
_
(dr)
2
+ r
2
(d)
2
.
Eliminating ds, we obtain
2dr =
_
(dr)
2
+ r
2
(d)
2
7
or
dr
r
=
d

3
,
and so
r() = C exp(/

3).
This path is a logarithmic spiral. Thus the best strategy for the patrol craft consists of
two steps: (1) proceed in a straight line to some point (r
0
,
0
) such that the craft and the
sub are the same distance r
0
from the origin; (2) follow the spiral r = r
0
exp[(
0
)/

3].
The craft will catch the sub at some point during the 360

(= 2) traversal on the
spiral.
1.3 Linearisation
Can we learn anything about the solution of the dierential equation
x = f(x)
if f(x) is a complicated nonlinear function of x? Suppose we are interested in the solution
x(t) only for x near x(0) = x
0
. Then we only need to know how f(x) behaves near x
0
.
Example. The nonlinear equation
x = ax + bx
2
is not easy to integrate. However, if bx
2
is small compared with ax, we have approximately
x = ax. On integrating this simpler linear equation, we obtain an approximate solution
x(t) = x(0) exp(at).
In general, to simplify a nonlinear equation x = f(x), we can use Taylors theorem
and approximate f(x) by the rst two terms of its Taylor expansion about x
0
:
f(x) f(x
0
) + f

(x)|
x=x
0
(x x
0
).
Introducing (t) = x(t) x
0
, we have approximately

= a + b, (0) = 0,
where
a = f

(x)|
x=x
0
= f

(x
0
), b = f(x
0
).
Assuming that a = 0, we integrate the linear equation for (t) to get
(t) =
b
a
e
at

b
a
,
and so, approximately,
x(t) = x
0
+ (t)
_
x(0)
b
a
_
+
b
a
e
at
.
8
Example. Suppose we are interested in the solution of
x = 1 sin[ln(1 +x)], x(0) = 0
only for x near x(0) = 0. We replace the right-hand side of the equation by the linear
Taylor approximation of f(x) = 1 sin[ln(1 + x)] about x = 0. We have f

(x) =
cos[ln(1 + x)]/(1 + x) and f

(0) = 1, so that approximately x = 1 x with the


general solution x(t) = 1 + C exp(t). The solution that satises the initial condition is
x(t) = 1 exp(t).
Thus we simplied a problem by approximating a complicated function f(x) with a
simpler linear function, dened by the rst two terms in the Taylor series of f(x). We
say that we linearised the original nonlinear dierential equation. The solution of the
linearised equation can often provide a good approximation to the exact solution x(t), at
least for suciently small times. We basically gained some knowledge at the expense of
some accuracy. This idea is often used in applied mathematics. Later we shall repeatedly
use linearisation to study stability of equilibrium solutions to dierential equations.
1.4 Separable and linear equations
Generally speaking, it is dicult to solve ordinary dierential equations. We consider
two types of rst-order equations for which routine methods of solution are available:
separable and linear rst-order equations.
A rst-order dierential equation is separable if it can be written in the form
x =
f(t)
g(x)
.
To solve this, we have only to write it in the form g(x)dx = f(t)dt and integrate:
_
g(x)dx =
_
f(t)dt + C.
Example. As we noted earlier, exponential growth cannot continue indenitely in the real
world. P. F. Verhulst (1838) suggested an extremely important modelthe logistic growth
modelthat describes the limiting eect of nite resources on the size of a population:

N = a
_
1
N
K
_
N,
where a and K are positive constants, and K is called the carrying capacity. If K is
innity large, we have

N = aN and the population size N(t) is increasing exponentially,
N(t) = N(0) exp(at). What is the eect of a nite K? The logistic equation is separable:
KdN
N(K N)
= adt.
9
We can use the change of variables x = 1/N to integrate the left-hand side:
_
KdN
N(K N)
=
_
Kdx
Kx 1
= ln(Kx 1) + lnC.
It follows that
K
N
1 = Ce
at
or
N(t) =
K
1 +C exp(at)
.
The initial condition at t = 0 species the integration constant:
K
N(0)
1 = C,
which yields
N(t) =
KN(0)
N(0) + (K N(0)) exp(at)
.
We see that, for any N(0), the exponential growth is limited:
N(t )
KN(0)
N(0)
= K.
Thus the population can only reach a nite size equal to the carrying capacity K, which
explains the meaning of the term.
The logistic growth model also describes the spread of technological innovations, say
substitution of electric locomotives for steam locomotives or adoption of high-speed bottle
llers in the brewing industry.
Another important type of dierential equation is the linear equation. The general
linear rst-order equation is
x = a(t)x + b(t).
If the equation is homogeneous (b = 0), it is separable, and its solution is
x(t) = x(0) exp
__
t
0
a(s)ds
_
.
Now we try to generalise this solution to solve a harder inhomogeneous equation. If b = 0,
we search for a solution in the form
x(t) = y(t) exp
__
t
a(s)ds
_
,
where y(t) is a new unknown function. The method works because the equation for y(t)
turns out to be simpler than that for x(t). We have
x = y exp
__
t
a(s)ds
_
+ y exp
__
t
a(s)ds
_
d
dt
__
t
a(s)ds
_
= y exp
__
t
a(s)ds
_
+ a(t)x,
10
and substitution back into the original dierential equation gives
y exp
__
t
a(s)ds
_
= b(t).
This is a separable equation. Its solution is
y(t) =
_
b(t) exp
_

_
t
a(s)ds
_
dt + C,
which yields the general solution
x(t) =
__
b(t) exp
_

_
t
a(s)ds
_
dt + C
_
exp
__
t
a(s)ds
_
.
Example. Oxygen concentration in a lake, n(t), may be described by
n = k(n
e
n) + S, n(0) = n
0
,
where k is a positive constant, n
e
is an equilibrium concentration in the absence of exter-
nal sources and sinks of oxygen (say, due to photosynthesis or pollution), and S(t) is a
function describing the sources and sinks. Suppose that S(t) = t. To solve the resulting
inhomogeneous linear equation,
n + kn = kn
e
+ t,
we seek a solution in the form n(t) = f(t)e
kt
. Substituting this into the equation yields
a separable equation for the new function f(t):

f = (kn
e
+ t)e
kt
.
Using integration by parts, we get
f(t) =
_
(kn
e
+ t)e
kt
dt
=
1
k
(kn
e
+ t)e
kt


k
_
e
kt
dt
=
_
n
e
+

k
t

k
2
_
e
kt
+ C.
As usual, the initial condition species C, and we nally arrive at
n(t) = n
e


k
2
+

k
t +
_
n
0
n
e
+

k
2
_
e
kt
.
Although solutions of nonlinear dierential equations are not easy to obtain, some
nonlinear equations can be solved by reducing them to simpler equations. For example,
Bernoullis equation
x = f(t)x + g(t)x

11
is nonlinear unless = 0 or = 1. The logistic growth model corresponds to = 2. For
an arbitrary , we can reduce Bernoullis equation to a linear equation by the change of
variable y = x
1
. We have
y = (1 )x

x = (1 )fx
1
+ (1 )g,
which yields a linear equation for y(t):
y = ay + b, a = (1 )f, b = (1 )g.
Example. An equation that generalises the logistic growth model is

N = a
_
1
_
N
K
_
s
_
N,
where a, K, and s are constants. This nonlinear equation is Bernoullis equation with
= s + 1. On substituting y(t) = N
s
, we obtain the linear equation
y = say + saK
s
with the general solution
y(t) = N
s
= K
s
+ Ce
sat
or
N(t) =
K
[1 +CK
s
exp(sat)]
1/s
.
We see that, as long as sa > 0, N K as t . The initial condition N(0) = N
0
species the integration constant C = N
s
0
K
s
, and so
N(t) = K
_
1 +
_
_
K
N
0
_
s
1
_
e
sat
_
1/s
.
This model was used to describe the growth of pine trees in New Zealand (O. Garcia,
1983).
1.5 Equilibrium and stability
Equilibrium solutions to dierential equations are dened as solutions that do not depend
on time: x = 0. We are often interested in equilibrium solutions. For example, we
may want to determine the conditions required for sustainable harvesting of a renewable
resource (say, shing) or an equilibrium of supply and demand in economics. To model
the global climate and climate change, we rst need to describe an equilibrium between
the solar radiation the Earth absorbs and the infrared radiation it emits.
For a rst-order equation of the form
x = f(x),
12
we nd all equilibrium solutions x = x
0
by solving
f(x
0
) = 0.
An equilibrium x = x
0
is called stable if every solution x(t) that starts near x
0
remains
near x
0
and x(t) x
0
as t . Otherwise, an equilibrium is called unstable.
We are usually interested in stable equilibrium solutions. Nature is full of perturba-
tions that destroy unstable equilibria, so they are rarely seen in natural systems.
Suppose we have x = f(x) and an equilibrium solution x = x
0
satises f(x
0
) = 0.
To investigate its stability, we need to determine the behaviour of a solution x(t) that is
initially very close to x
0
. Hence we only need to know how f(x) behaves near x
0
. Using
the rst two terms of the Taylor expansion about x
0
, we get
f(x) f(x
0
) + f

(x
0
)(x x
0
) = f

(x
0
)(x x
0
),
and we can approximate the original dierential equation by
x = f

(x
0
)(x x
0
),
which yields
x(t) = x
0
+ C exp(f

(x
0
)t).
Clearly x(t) x
0
for large t, provided f

(x
0
) < 0, Conversely, x(t) deviates strongly from
x
0
for large t, if f

(x
0
) > 0. Thus we have the following stability criterion: if x = f(x) and
f(x
0
) = 0, then x = x
0
is a stable equilibrium if f

(x
0
) < 0 and an unstable equilibrium
if f

(x
0
) > 0.
Example. The equilibrium solution x
0
= 1 to the equation x = ln x is unstable because
f

(x) = d lnx/dx = x
1
, and therefore f

(x
0
) = x
1
0
= 1 > 0.
Example. Suppose water ows into a lake at a constant rate r and that water evaporates
from the lake at a rate that is proportional to (volume)
2/3
. We have the following equation
for the volume V of the lake:

V = f(V ) = r aV
2/3
,
where a is a constant. We solve f(V
0
) = 0 to obtain an equilibrium solution:
V
0
=
_
r
a
_
3/2
.
To determine its stability, we calculate f

(V ) =
2
3
aV
1/3
. Now
f

(V
0
) =
2
3
aV
1/3
0
=
2
3
a
3/2
r
1/2
,
and we conclude that the equilibrium is stable because f

(V
0
) < 0.
13
Example. The logistic equation

N = f(N) = a
_
1
N
K
_
N
has two equilibrium solutions: N
0
= 0 and N
0
= K. Note that we can assume a = 1 and
K = 1 without loss of generality. Indeed, using the variables N

= N/K and t

= at, we
get
dN

dt

= N

(1 N

),
which is the logistic equation with a = K = 1 for the new variables. So, assuming that
a = K = 1, we have the equilibrium solutions N
0
= 0 and N
0
= 1. Now,
f(N) = N(1 N), f

(N) = 1 2N,
and we conclude that N
0
= 0 is unstable since f

(0) = 1 > 0, whereas N


0
= 1 is stable
since f

(1) = 1 < 0. This is consistent, of course, with the exact solution to the logistic
equation.
Example. The same mathematical equation can arise in very dierent contexts. Consider
a model for HIV spread among drug users in a large housing complex. The total number
of users N consists of I infected and S susceptible (uninfected) individuals. We assume
that the rate of HIV transmission is proportional to the number of encounters between
the infected and susceptible users, SI, and that the death rate is proportional to the
number of infected, I, where and are proportionality constants. We also assume
that the total population N = I +S does not change (as tenants die, new ones move in).
Then we have

I = SI I
= (N I)I I
=
_
N

I
_
I,
which we recognise as the logistic equation in dierent notation. If N > /, the only
stable equilibrium solution
I
0
= N

> 0,
which means that the disease is constantly present in the population. By contrast, if
N < /, the solution I
0
< 0 can be discarded as meaningless, and the other equilibrium
solution I = 0 is now stable, which means that the disease disappears from the population.
Hence the model suggests that the disease will be eradicated when N < /. This can be
accomplished by either of the following three methods: (1) decrease the population size
N, say by resettling the tenants in smaller housing units; (2) decrease the transmission
rate , say by establishing a needle exchange program; (3) increase the death rate , say
by withholding treatment. Clearly not all mathematically correct solutions are socially
acceptable.
14
Example. If a sh population is growing logistically and is being harvested at a constant
rate, the population size N(t) is described by

N = f(N) = (1 N)N H, H = const.


Here N and t are measured in such units that a = K = 1, and H is the sh harvesting
rate (the total catch per unit time) that we want to maximise. We describe a sustainable
shery by an equilibrium solution N
0
:
(1 N
0
)N
0
H = 0
or
N
0
=
1
2
_
1

1 4H
_
,
hence H 1/4. Now f

(N) = 1 2N, and we see that N


0
= (1

1 4H)/2 is
unstable because f

(N
0
) =

1 4H > 0, whereas N
0
= (1 +

1 4H)/2 is stable
because f

(N
0
) =

1 4H < 0. When H = 1/4, the two equilibrium solutions merge


into N
0
= 1/2. It is tempting to choose H = 1/4 as the maximum sustainable harvesting
rate, but is the corresponding population N
0
= 1/2 stable? We cannot use our stability
criterion because f

(1/2) = 0. We can integrate the dierential equation though. When


H = 1/4, we have

N = (1 N)N
1
4
=
_
N
1
2
_
2
,
and on separating variables and integrating, we get
1
N
1
2

1
N(0)
1
2
= t.
For the initial condition
N(0) =
1
2
, > 0
we have
N(t) =
1
2


1 t
.
If is small enough, we have N(0) that is very close to N
0
= 1/2. Yet N(t) does not
remain near N
0
. Therefore the equilibrium N
0
= 1/2 is unstable. In fact N(t) = 0 when
t =
1
2, which means that the maximum harvesting rate H = 1/4 will lead to the
collapse of a shery.
Example. We can prevent the collapse described in the previous example by introducing
a feedback into the model:

N = f(N) = (1 N)N H, H = N.
15
Now the harvesting rate will decrease or increase depending on the current population
size, which will have a stabilising eect on the population. Indeed, equilibrium solutions
follow from
(1 N
0
)N
0
N
0
= 0,
which yields N
0
= 0 and N
0
= 1 . Clearly we obtain a positive N
0
if < 1. Since
f

(N) = 1 2N, we see that N


0
= 0 is unstable because f

(N
0
) = 1 > 0, whereas
N
0
= 1 is stable because f

(N
0
) = 1 < 0. We can express the harvesting rate H
as a function of :
H = N
0
= (1 ).
We know from calculus that we can determine the maximum of a function by calculating
its derivative and equating the derivative to zero. We obtain
dH
d
= 1 2 = 0.
Hence the sustainable harvesting rate H is maximised by choosing = 1/2. Surprisingly,
although the equilibrium is stable, the maximum H = (1 ) = 1/4 is the same as that
for the unstable equilibrium of the previous example.
Exercises
1. Half of a spherical snowball melts in an hour. How long will it take for the remainder
to melt? Assume that the snowball remains a sphere at all times and that its volume
decreases at a rate proportional to its surface area.
2. Charcoal from a cave at an archaeological site is characterised by an average of 0.97
disintegrations of radioactive carbon-14 per minute per gram. Living wood is characterized
by an average of 6.68 disintegrations of carbon-14 per minute per gram. The half-life of
carbon-14 is approximately 5700 years. Estimate the date of occupation of the cave.
3. In an economic model, the price p of a commodity satises the equation
p = (D Q),
where the demand and supply are given by D(p) = a +bp and Q(p) = c +dp, and , a, b,
c, and d are constants. You may assume > 0, b < 0, d > 0. Determine p(t) if p
0
is the
commodity price at t = 0. Use the solution to determine the limit of p(t) as t .
4. A new lecturer begins checking books out of the university library at the rate of one
per day. Every week, he returns to the library 1/10 of the total number of books N(t)
checked out. Assuming that N(t) can be approximated by a continuous function of time
t, write down a dierential equation for N(t). Specify a reasonable initial condition N(0)
and solve the dierential equation. How many library books does the lecturer have at any
time after he has been around for a few years?
16
5. Suppose a quantity a x of a substance A is present and changes into a substance B
at a rate k
1
(a x), while the substance B is changing into the substance A at a rate k
2
x.
Determine x(t), assuming x(0) = 0.
6. The number of births per unit time in a population x(t) of zebras in a game park is
Bx. The number of zebras that die per unit time is L + Dx where L is the number of
deaths due to lions and Dx those due to other causes. Assume that B, L, and D are
known constants (B = D). If x
0
is the population at t = 0, nd x(t) for t > 0.
7. The motion of a body falling in a viscous medium may be described by the equation
m v = bv mg,
where m, g, b are positive constants. Given the initial conditions v(0) = 0 and x(0) = 0,
determine the velocity v(t) and displacement x(t) for t 0.
8. The size of a population is described by

N =
N
0
2
sin
_
N
2N
0
_
, N(t
0
) = N
0
.
Determine N(t). What is the limit of N(t) as t ?
9. The population size N(t) satises the dierential equation

N = kN
2
,
where k is a constant. Determine N(t) for t 0 if N(0) = N
0
. Sketch the solution if
k > 0 (explosive growth) and if k < 0 (catastrophic decay).
10. The seasonal growth of a tree is modelled by
y = (1 + sin t)y, y(0) = y
0
,
where and are positive constants. Determine y(t).
11. A certain toxin destroys a strain of bacteria at a rate proportional to the product
of the number of bacteria, N(t), and the amount of toxin present, T(t). If there were no
toxin present, the bacteria would grow at a rate proportional to N(t). Solve the resulting
ordinary dierential equation for the number of bacteria at time t:

N = (a T)N,
assuming that T(t) = bt and N(0) = N
0
, where a and b are known constants. Use the
solution to determine the limit of N(t) as t .
12. The height h(t) of a growing person is modelled by

h = a(t)(h
1
h), h(0) = h
0
,
17
a(t) = a
0
+ a
1
t,
where a
0
, a
1
, h
0
, h
1
are positive constants, and h
0
< h
1
. Determine h(t). What is the
limit of h(t) as t ?
13. A global warming model leads to the following equation for the average temperature
of the Earths surface:

T = k
1
exp(k
2
t) k
3
(T T
0
),
where k
1
, k
2
, k
3
are positive constants. Determine T(t), assuming that T() = T
0
.
14. The rate of a chemical reaction A + B AB is proportional to the product of the
concentrations of A and B. At the same time, the compound AB itself breaks apart into
A and B at a rate proportional to its own concentration x(t). The resulting equation for
x(t) < 1 in appropriate units is as follows:
x = (1 x)
2
kx.
Derive the equilibrium solution x = x
0
and determine whether it is stable.
15. The size N(t) of an insect population satises the dierential equation

N = bN
2
aN.
Here a and b are known positive constants. Derive an equilibrium solution N
0
> 0 and
determine whether it is stable.
16. Consider a circular colony of bugs on a plate. If N is the total number of bugs, the
Malthusian growth rate is given by r
1
N. Bugs on the perimeter, however, suer from cold
and they die at a rate r
2
N
1/2
. Hence the dierential equation satised by N is

N = r
1
N r
2
N
1/2
.
Is there an equilibrium solution N
0
> 0? If so, is it stable?
17. Your lecturer would like to lose some weight. He begins by constructing a mathemat-
ical model for dieting in which his weight W(t) changes due to food intake F(t), exercise
E(t), and body metabolism M(t):

W = F E M, M = bW
3/4
.
Assume that F E and b are positive constants. Derive an equilibrium solution W
0
and
determine whether it is stable.
18
Chapter 2
Second-order ordinary dierential
equations
2.1 Linear second-order equations
The order of a dierential equation is the order of the highest derivative in the equation.
A second-order equation for a function x(t) is an equation of the form
x = f(t, x, x).
The notation x = d
2
x/dt
2
is used when the independent variable t is time.
Second-order dierential equations arise quite often in applications. The most famous
example is Newtons second law of motion
m x = F(t, x, x),
which governs the motion of a particle of mass m along the x-axis under the inuence of
a force F.
Second-order dierential equations are much more dicult to solve than rst-order
equations. Fortunately, many second-order equations of practical interest are linear. We
only consider linear second-order dierential equations:
x + p x + qx = f,
where p(t), q(t), and f(t) can only depend on the independent variable t. No loss of
generality results from taking the coecient of x to be 1 since this can always be accom-
plished by division. If f = 0, the equation is called inhomogeneous (or nonhomogeneous).
If f = 0, it is homogeneous.
Some theoretical questions about the existence and nature of solutions of dierential
solutions are very dicult. In applications, however, we typically expect that the general
solution to a second-order equation contains two arbitrary constants that can be specied
by initial conditions. For example, we can use the values of the function itself and its rst
derivative at a given time.
19
Example. Motion with a constant speed v
0
corresponds to zero acceleration: x = 0. On
integrating twice, we obtain the general solution to this simple second-order equation:
x = C
1
, x(t) =
_
C
1
dt = C
1
t + C
2
,
where C
1
and C
2
are constants of integration. We use the initial conditions x(0) = x
0
and x(0) = v
0
to obtain x(t) = x
0
+ v
0
t.
Example. Suppose the temperature T(x) varies along a straight line (the x-axis) and
does not depend on time. It turns out that in this case the heat equation simplies to
T

(x) = 0. The general solution is T(x) = C


1
x + C
2
. For instance, if T(0) = T
i
is the
temperature inside a house, T(a) = T
o
is the temperature outside, and a is the thickness
of a window pane, then the temperature prole inside the window pane is
T(x) = T
i
+ (T
o
T
i
)
x
a
, 0 x a.
We have specied a unique solution by using the known values of T at two dierent points
(boundaries): x = 0 and x = a. Problems of this type are called boundary value problems.
Now we can calculate the rate of heat loss through the window, dened as
Q = A
dT
dx
= A
T
i
T
o
a
.
Here is the coecient of thermal conductivity of glass, and A is the area of the window
pane. We see that Q

(x) = 0, and so the equation T

(x) = 0 means that the heat transfer


rate Q is the same through each cross-section x = const. The thermal conductivity of air
is about 16 times less than the thermal conductivity of glass. This is why the air gap in
a double glazed window considerably reduces the heat loss through the window.
Example. Second-order equations often arise in economics. Suppose z is the annual
demand for goods and services in a national economy:
z(t) = c(t) + x(t) + g(t),
where c is the annual amount of consumption, x is the annual amount of investment, and
g is the annual government expenditure. The so-called multiplier-accelerator model for
the annual supply y(t) of those goods and services assumes that
y = (z y), = const
and
c = (1 s)y, s = const,
where 0 < s < 1 and 1/s is termed the multiplier. Combining these equations, we have
y = [(1 s)y + x + g y] = (x + g sy).
20
Hence
x =
1

y + sy g,
which yields
x =
1

y + s y g.
The nal assumption of the model is that
x = k(v y x), k = const, v = const,
where v is termed the accelerator. Eliminating x and x in this equation, we get
y + s y g = k(v y
1

y sy + g),
or
y + p y + qy = f,
where p = (s kv) + k, q = ks, and f(t) = (kg + g). The model can describe a rich
variety of types of behaviour, depending on the parameters p and q and the government
policy, specied by f(t).
2.2 The homogeneous equation with constant coe-
cients
We can determine a complete solution of the homogeneous linear second-order equation
x + p x + qx = 0
for the special case in which p and q are constants. We begin by noting that the derivatives
of the exponential function x(t) = Aexp(t) are constant multiples of the function itself:
x = Ae
t
, x =
2
Ae
t
.
We try to nd the constant that makes x(t) = Aexp(t) a solution of the dierential
equation. On substituting this x(t) into the equation, we get
(
2
+ p + q)Ae
t
= 0.
We see that x(t) = Aexp(t) is a nonzero solution if
2
+p +q = 0. This is a quadratic
equation, so there are two solutions:

1
=
p
2
+

p
2
4q
2
,
2
=
p
2

p
2
4q
2
.
Thus x(t) = Aexp(
1
t) and x(t) = Bexp(
2
t) are both solutions, whatever the values of
the constants A and B, and the general solution is
x(t) = Ae

1
t
+ Be

2
t
.
21
If p
2
> 4q,
1
and
2
are distinct real numbers, and the general solution x(t) contains
two exponential functions.
Example. The equation x x = 0 is a linear homogeneous equation with constant
coecients: p = 0, q = 1. Because p
2
4q = 4 > 0, we have
1,2
= 1 and therefore
x(t) = Ae
t
+ Be
t
.
If p
2
< 4q, then
1
and
2
are complex numbers. If we denote =

4q p
2
/2, we
have
x(t) = Ae
pt/2+it
+ Be
pt/2it
= e
pt/2
_
Ae
it
+ Be
it
_
.
To specify a real-valued solution, we need Eulers formula (L. Euler, 1748). We derive it
formally by calculating
d
dt
_
e
it
(cos t + i sin t)
_
= ie
it
(cos t + i sin t) + e
it
(sin t + icos t)
= e
it
(i cos t + sin t sin t + i cos t)
= 0.
Hence exp(it)(cos t + i sin t) is a constant. We nd its value by substituting t = 0,
which yields exp(it)(cos t + i sin t) = 1. On multiplying either side by exp(it), we
obtain Eulers formula
e
it
= cos t + i sin t,
and similarly (replacing by )
e
it
= cos t i sin t.
Using Eulers formula in the general solution x(t), we see that
x(t) = e
pt/2
[(A+ B) cos t + i(A B) sin t] .
To ensure that x(t) is real, we choose A and B such that A+B = C is real and AB =
iD is imaginary. Remembering the denition of , we nally obtain a real-valued
solution
x(t) = e
pt/2
_
C cos

4q p
2
2
t + Dsin

4q p
2
2
t
_
.
Example. The equation x + x = 0 is a linear homogeneous equation with constant
coecients: p = 0, q = 1. Because p
2
4q = 4 < 0, we have = 1 and therefore
x(t) = C cos t + Dsin t.
22
A snag arises if p
2
= 4q. Because
1
=
2
, x(t) = Aexp(pt/2) contains only one
arbitrary constant, so it cannot be the general solution. We nd the general solution by
searching for it in the form x(t) = y(t) exp(pt/2). We have
x =
_
y
p
2
y
_
e
pt/2
, x =
_
y p y +
p
2
4
y
_
e
pt/2
,
and substitution into x + p x + qx = 0 yields
y +
_
q
p
2
4
_
y = 0
or
y = 0, y(t) = A + Bt.
Therefore,
x(t) = (A+ Bt)e
pt/2
is the general solution in the case p
2
= 4q.
Example. The equation x+4 x+4x = 0 is a linear homogeneous equation with constant
coecients: p = 4, q = 4. Because p
2
4q = 0, we have
1,2
= 2 and therefore
x(t) = (A + Bt) exp(2t). If initial conditions are given by x(0) = 1 and x(0) = 3, for
example, then we have 1 = A and 3 = B 2A, and nally x(t) = (1 + 5t) exp(2t).
Example. L. F. Richardson (1939) suggested a model that describes the relation between
two nations, each determined to defend itself against a possible attack by the other.
Suppose x(t) is the war potential (armaments) of nation 1 and y(t) is the war potential
of nation 2. The rate of change of x(t) depends on the war readiness y(t). In the simplest
model, we represent this rate by ky where k > 0 is a constant. Similarly, we model the
rate of change of y(t) by lx where l > 0 is a constant. Neglecting all other factors such
as the cost of armaments, we write
x = ky, y = lx.
Dierentiating the rst equation with respect to t and using the second equation to
eliminate y, we have
x klx = 0.
We solve this linear homogeneous equation to get
x(t) = Ae

klt
+ Be

klt
.
It follows that
y(t) =
x
k
=

l
k
_
Ae

klt
Be

klt
_
.
The integration constants A and B are determined by initial conditions. If A > 0, both
x(t) and y(t) tend to innity. This can be interpreted as war.
23
Example. Suppose now that two armies are engaged in combat, with x(t) and y(t)
denoting the number of combatants of the x and y forces. Who wins the battle? F. W.
Lanchester (1916) tried to answer this question using the following model. The rate of
change of x(t) depends on the ghting eectiveness of y, which we assume to be given by
by, operational losses (accidents, desertions), which we assume to be given by ax, and
reinforcements P(t). These assumptions enable us to write
x = ax by + P(t), x(0) = x
0
.
Similarly for the y-force,
y = cx dy + Q(t), y(0) = y
0
.
If the operational losses and reinforcements are negligible (a = d = P = Q = 0), we have
x = by, x(0) = x
0
,
y = cx, y(0) = y
0
.
Note that we must also have x(0) = by
0
. Now, on dierentiating the equation for x and
using the second equation to eliminate y, we get
x bcx = 0,
and the general solution is
x(t) = Ae

bct
+ Be

bct
.
The initial conditions x(0) = x
0
, x(0) = by
0
yield
x(t) = x
0
cosh(

bct)

b
c
y
0
sinh(

bct),
and similarly
y(t) = y
0
cosh(

bct)
_
c
b
x
0
sinh(

bct).
We can eliminate the explicit t-dependence by dividing one dierential equation by the
other:
dy
dx
=
y
x
=
cx
by
.
This is a separable rst-order equation. We integrate it to obtain an interesting result:
by
2
cx
2
= by
2
0
cx
2
0
= const.
Since this quantity never changes sign, only one of x and y can ever be zero. Therefore,
the y-force wins (y > 0 when x = 0) if by
2
0
cx
2
0
> 0, and the x-force wins (x > 0 when
y = 0) if by
2
0
cx
2
0
< 0. If a 50,000 y-force is ghting a 70,000 x-force with an equal troop
eectiveness, b = c, the larger x-force obviously wins. Suppose, however, that the y-force
meets sequentially x-armies of 40,000 and 30,000. The rst battle results in a y-force
victory with a margin of

50, 000
2
40, 000
2
= 30, 000, which is just enough to force a
draw with the second x-army. This illustrates the importance of concentration in tactics.
24
2.3 Linear oscillations
We often need to describe systems that exhibit periodic, oscillatory behaviour, for instance
the oscillations of a quartz crystal in a watch, the beating of the heart, the periodic rise
and fall of the ocean level due to tides, or business cycles in economics.
The simplest oscillatory system consists of a mass m attached to a spring, which
moves without friction in a straight line. If x denotes the displacement of the mass from
its equilibrium position and the spring exerts the restoring force F = kx (k > 0) on m,
then the equation of motion m x = F is
x +
2
x = 0,
2
=
k
m
with the general solution
x(t) = Acos t + Bsin t.
Here is called the circular frequency, and the integration constants A and B are specied
by initial conditions.
Example. If x(0) = 2 and x(0) = 0, then A = 2, B = 0, and so x(t) = 2 cos t.
If we dene an amplitude a
0
=

A
2
+ B
2
and a phase
0
= arctan(A/B) of the
oscillations, we can write the integration constants as A = a
0
sin
0
and B = a
0
cos
0
.
Hence we can express the general solution as follows:
x(t) = a
0
sin(t +
0
).
This form of the solution makes it clear that the oscillations are indeed periodic, x(t+T) =
x(t), with the period
T =
2

= 2
_
m
k
.
Example. Suppose a spherical buoy of radius R is oating half submerged in water. We
have mg = F
0
in equilibrium, where m is the mass of the buoy, g is the acceleration due to
gravity, and F
0
is an expulsion (Archimedess) force equal to the weight of the displaced
water. If the buoy is depressed slightly and then released, a restoring force pushes it
upward, causing oscillations. If the friction of the water is neglected, the oscillations
are described by m x = F, where x(t) is the vertical displacement of the buoy, and
F = F F
0
is the change in the expulsion force. If the displacement is so small that
x R, we can calculate F = F
0
V/V , where V is the small change in the submerged
volume:
V
V
=
R
2
x
2R
3
/3
=
3x
2R
.
Hence
m x = F =
3x
2R
F
0
=
3x
2R
mg
or
x +
3g
2R
x = 0,
25
and so the period of the resulting oscillations is T = 2
_
2R/(3g).
Consider now the possibility that there is a friction force F
f
that will damp the oscil-
lations. We assume that the force is proportional to the speed of the oscillating mass:
F
f
= rv = r x, r > 0.
The equation of motion becomes
m x + r x + kx = 0
or, introducing = r/(2m) and
2
0
= k/m,
x + 2 x +
2
0
x = 0.
From now on,
2
0
= k/m denotes the natural frequency, that is the frequency at which
the system would oscillate if there were no friction force. As usual, we solve this linear
equation by assuming that x(t) = Aexp(t), which leads to

2
+ 2 +
2
0
= 0
and so

1,2
= i, =
_

2
0

2
.
We observe that three dierent types of solution are possible, depending on whether
<
0
, >
0
or =
0
.
If the damping is weak, <
0
, we say that the system is underdamped. In this case
the general solution is
x(t) = (Acos t + Bsin t)e
t
= a
0
sin(t +
0
)e
t
.
Thus the mass executes oscillations, but their amplitude decreases exponentially with
time. We dene the period T of the decaying oscillations as the time between successive
maxima of the function x(t):
T =
2

=
2
_

2
0

2
=
2
_
k/mr
2
/(4m
2
)
.
If the damping is strong, >
0
, we say that the system is overdamped. In this case

1
and
2
are real and both negative. Hence the general solution is
x(t) = Ae
|
1
|t
+ Be
|
2
|t
.
The friction force is so large that there are no oscillations. Instead x(t) decays exponen-
tially, so that the mass approaches the equilibrium position x = 0 as t . Depending
on initial conditions, the mass can cross x = 0 at most once.
In the mathematically interesting case of critical damping, =
0
, we have
x(t) = (A+ Bt)e
t
.
The behaviour of the solution is similar to its behaviour in the overdamped case since the
exponential decay overcomes an algebraic growth. From a physical point of view, however,
this case is insignicant because in practice it is very unlikely that =
0
exactly.
26
2.4 Equilibrium and stability
We are often interested in equilibrium solutions of second-order dierential equations,
which are solutions that do not depend on time: x = x
0
is an equilibrium solution if
x = x = 0. In mechanical problems, x = 0 means that all forces cancel, and x = 0 means
that there is no motion. For a second-order equation of the form
x = f(x, x),
we nd all equilibrium solutions x = x
0
by solving
f(x
0
, 0) = 0.
An equilibrium x = x
0
is called asymptotically stable if every solution x(t) that starts
near x
0
remains near x
0
and x(t) x
0
as t .
As for rst-order dierential equations, we can determine which equilibrium solutions
are asymptotically stable by using a linearised stability analysis. Suppose an equilibrium
solution x = x
0
satises f(x
0
, 0) = 0. To investigate the stability of x = x
0
, we need to
determine the behaviour of a solution x(t) that is initially very close to x
0
. Hence we
only need to know how f(x, x) behaves near x
0
. We expand f(x, x) in a Taylor series as
a function of two variables:
f(x, x) = f(x
0
, 0) +
f
x

x
0
,0
(x x
0
) +
f
x

x
0
,0
x + . . .

f
x

x
0
,0
(x x
0
) +
f
x

x
0
,0
x,
where we kept only the linear terms in the expansion and used f(x
0
, 0) = 0 and x
0
= 0.
If (t) = x x
0
denotes a deviation from the equilibrium solution, we have

=
f(x
0
+ ,

), or approximately

+ p

+ q = 0,
where
p =
f
x

x
0
,0
, q =
f
x

x
0
,0
.
This is a second-order homogeneous equation with constant coecients. Its general solu-
tion is (t) = Aexp(
1
t) + Bexp(
2
t) where as before

1,2
=
1
2
_
p
_
p
2
4q
_
.
We require for asymptotic stability that the deviation remain small and 0 as t .
Hence we require that the real parts of both
1
and
2
be negative. If q < p
2
/4, both
roots are real, and we require that the larger root be negative: (p +

p
2
4q)/2 < 0,
which is the case if q > 0. If q > p
2
/4, both roots are complex, but their real parts are
27
the same, and so we require that p/2 < 0, which is the case if p > 0. Summarising these
results, we obtain
p > 0, q > 0.
Thus we have the following stability criterion. If x = f(x, x) and f(x
0
, 0) = 0, then x = x
0
is a stable equilibrium solution if
f
x

x
0
,0
< 0,
f
x

x
0
,0
< 0.
The equilibrium solution is unstable if either of the two inequalities is not satised.
Example. The van der Pol equation
x + (x
2
1) x + x = 0, > 0
was encountered in the study of electric circuits (B. van der Pol, 1920). It is clear that
x
0
= 0 is the only equilibrium solution. Since f(x, x) = x (x
2
1) x, we have
f
x

x
0
,0
= 1,
f
x

x
0
,0
= (x
2
0
1) = .
Thus the equilibrium solution is unstable.
Sometimes other denitions of stability may be useful. For example, if (t) = x x
0
remains small but does not tend to zero as t , the equilibrium solution x
0
is said to
be neutrally stable.
Example. The only equilibrium solution of the linear oscillator equation
x +
2
x = 0
is x
0
= 0. Now f(x, x) =
2
x, and so (f/x)|
x
0
,0
=
2
< 0 but (f/ x)|
x
0
,0
= 0.
Thus the equilibrium is not asymptotically stable. It is neutrally stable, however, since
the general solution x(t) describes periodic oscillations about x
0
.
Generally, x = x
0
is a neutrally stable equilibrium solution of the equation x = f(x)
if f

(x
0
) < 0.
We can use linearisation to study asymptotic stability of equilibrium solutions to
x = f(x, x) and neutral stability of equilibrium solutions to x = f(x). When the context is
clear, we simply say that an equilibrium is either stable or unstable. It turns out, however,
that linearisation cannot be used to study neutral stability of equilibrium solutions to
x = f(x, x). A. M. Lyapunov (1892) discovered a powerful but more complicated method
for investigating stability in this case. A solution is said to be stable in the sense of
Lyapunov if it is either neutrally stable or asymptotically stable.
28
2.5 Inhomogeneous equations
We now consider a second-order inhomogeneous equation
x + p x + qx = f, f = 0
and suppose x
p
(t) is a particular solution of this equation, so that
x
p
+ p x
p
+ qx
p
= f.
On subtracting one equation from the other, we obtain a homogeneous equation
x
h
+ p x
h
+ qx
h
= 0
for the function x
h
(t) = x(t) x
p
(t). We see that the general solution x(t) of an inho-
mogeneous dierential equation is the sum of the general solution x
h
of the homogeneous
equation and a particular solution x
p
of the inhomogeneous equation:
x(t) = x
h
(t) + x
p
(t).
If both p and q are constants, we can determine a complete solution of the homogeneous
equation, and therefore we can write the general solution of the inhomogeneous equation
as
x(t) = x
h
(t) + x
p
(t) = Ae

1
t
+ Be

2
t
+ x
p
(t)
(suitably modied if
1
and
2
are complex numbers or if
1
=
2
). For example, if
f = const, we can simply choose x
p
= f/q = const.
More generally, if f(t) is a simple function, we can nd x
p
by the method of un-
determined coecients. The idea is to search for x
p
(t) in a form suggested by the
form of f(t). For instance, if f(t) = at
n
, we seek x
p
(t) = C
0
+ C
1
t + . . . + C
n
t
n
. If
f(t) = a exp(t), we seek x
p
(t) = C exp(t). If f(t) = a
1
cos(t) + a
2
sin(t), we seek
x
p
(t) = C
1
cos(t) +C
2
sin(t). We determine the coecients in x
p
(t) by substituting x
p
into the inhomogeneous dierential equation. We should be careful if f(t) = a exp(t)
and coincides with
1
or
2
. If =
1
but
1
=
2
, we seek x
p
(t) = Ct exp(
1
t). We
can justify this assumption by an argument similar to that used in the derivation of the
solution of the homogeneous equation in the case
1
=
2
. Finally, if =
1
=
2
, we
seek x
p
(t) = Ct
2
exp(
1
t).
Example. A linear frictionless oscillator, driven by a periodic external force, is described
by
x +
2
0
x = sin t.
Consider rst the case =
0
. We try x
p
(t) = C sin t. On substituting this into the
equation, we get
C
2
sin t + C
2
0
sin t = sin t,
which can be satised only if C = 1/(
2
0

2
). Thus we have the general solution
x(t) = Acos
0
t + Bsin
0
t +
1

2
0

2
sin t.
29
If the initial conditions are given by x(0) = 0 and x(0) = 0, for example, we have
A = 0, B
0
+

2
0

2
= 0,
and the solution is
x(t) =
1

2
0
_

0
sin
0
t sin t
_
.
Example. Now suppose the frequency of the external force is equal to the natural
frequency of the system. We cannot just substitute =
0
into the above solution for
x(t) since that would lead to division by zero. We can, however, dene the solution in
this case as a limit:
x(t) = lim

0
(/
0
) sin
0
t sin t

2
0
.
On using LHospitals rule to evaluate the limit, we nd
x(t) =
d
d
[(/
0
) sin
0
t sin t]
d
d
(
2

2
0
)

=
0
=
1
2
2
0
sin
0
t
t
2
0
cos
0
t.
Alternatively, we could nd this solution using the method of undetermined coecients.
When =
0
, seek a particular solution
x
p
(t) = C
1
t cos(
0
t) + C
2
t sin(
0
t)
and determine the constants C
1
= 1/(2
0
) and C
2
= 0 by substituting x
p
(t) into the
dierential equation. It follows that the general solution is
x(t) = x
h
(t) + x
p
(t) = Acos
0
t + Bsin
0
t
t
2
0
cos
0
t.
As before, the integration constants A and B are specied by the initial conditions. We
see that even though the external force is periodic, the amplitude of the solution grows
with time (strictly speaking, only until a friction force is no longer non-negligible).
The phenomenon of a strong response of an oscillator when driven at the right fre-
quency is called resonance. An impressive example of resonance is the vocal coach Jaime
Vendera holding a note and shattering a wine glass when the frequency of the note is
equal to the natural frequency of the glass. An everyday example of resonance is tuning.
When we tune a radio, we change the parameters of an electric circuit to match the carrier
frequency of a particular station.
Example. Consider an electric circuit consisting of an inductor of inductance L, a resistor
of resistance R, a capacitor of capacitance C, and a battery producing an electromotive
30
force (voltage) E(t), all connected in series. The governing equation for the electric current
I(t) in the circuit is
L

I + RI +
Q
C
= E,
where Q is the electric charge. On dierentiating this equation and using I =

Q, we have
L

I + R

I +
I
C
=

E.
For example, if E = E
0
sin(t +) where E
0
and are constants, the general solution of
this inhomogeneous equation is
I(t) = I
h
(t) + I
p
(t).
Here the general solution of the homogeneous equation is
I
h
(t) = Ae
Rt/(2L)
sin
_
4CL R
2
C
2
2CL
t +
_
,
where A and are integration constants, and we assume the underdamped case R <
2
_
L/C. A particular solution of the inhomogeneous equation is
I
p
(t) =
CE
0
[RC sin(t + ) + (1 CL
2
) cos(t + )]
(RC)
2
+ (1 CL
2
)
2
.
The homogeneous solution I
h
decays with time, so I
h
is called the transient current,
whereas I
p
is called the steady state current. Resonance occurs if = 1/

CL, in which
case the amplitude of I
p
reaches the maximum value of E
0
/R. Note that the solution
remains bounded as long as the resistance R is nonzero, but its amplitude will be very
large if R is very small.
Exercises
1. Solve the following ordinary dierential equation:
u

+ u

+ u = 1.
Comment: the problem appears in Heard on the Street: Quantitative Questions from
Wall Street Job Interviews by T. F. Crack.
2. Radium decays to radon which decays to polonium. This two-step process of radioac-
tive decay is described by the following equations:

N
1
=
1
N
1
, N
1
(0) = N
0
,

N
2
=
1
N
1

2
N
2
, N
2
(0) = 0.
31
Derive a second-order dierential equation with constant coecients for N
2
(t). Determine
N
1
(t) and N
2
(t). Investigate separately the cases
1
=
2
and
1
=
2
.
3. A particle slides without friction in a long slender tube that rotates in a vertical plane
about its centre with constant angular frequency . If the tube is horizontal at time t = 0
and g > 0 is a constant gravitational acceleration, then the particle displacement u(t)
from the centre is governed by
u
2
u = g sin t.
Find the general solution u(t) and solve the initial value problem u(0) = 0, 2 u(0) = g.
4. Imagine a tunnel connecting two points on the surface of the Earth, which passes
through the centre of the Earth. Neglecting air resistance, the only force acting on a
body of mass m in the tunnel is the gravitational force: F = mgr/R. Here g 9.8
m s
2
is the gravitational acceleration at the surface, r is the distance from the Earths
centre, and R 6400 km is the radius of the Earth. Write down a second-order dierential
equation that describes the motion of a body in the tunnel. Determine the general solution
r(t) to the equation and solve the initial value problem r(0) = R, r(0) = 0. For these
initial conditions, calculate the time it takes for the body to travel between the two ends
of the tunnel and the maximum speed reached in the course of the travel.
5. Determine whether x
0
= 0 is a stable solution of the following equation:
x + 6 x 4x = 0.
6. Suppose nonlinear oscillations are described by
x + x = x x
2
.
Determine all equilibrium solutions and investigate their stability.
7. A mathematically-minded vandal wishes to break a hot-water radiator of mass m
away from its foundation, but nds that he cannot achieve this by applying steadily the
greatest force of which he is capable. He nds, however, that he can apply a periodic
force f(t) = f
0
sin(t). The foundation of the radiator resists its movement by a force
proportional to its displacement. The resulting dierential equation for the displacement
x(t) is as follows:
x +
k
m
x =
f(t)
m
.
Assuming that the frequency is chosen so that k/m =
2
, determine x(t) for the initial
conditions x(0) = 0, x(0) = 0. Will the vandal achieve his goal?
32
Chapter 3
Elements of classical mechanics
3.1 Kinematics
Classical mechanics involves solving Newtons equations of motion to explain and predict
how various objects movehow projectiles and aircraft move through the air, how space-
craft move around the Earth, how planets move around the Sun. Kinematics describes
the motion of objects, without reference to the forces causing the motion. Dynamics deals
with the forces that produce motion. Numerous applications of calculus arise in classical
mechanics.
When the size of a moving object is small enough, we can treat the object as a point
mass, or particle. A particle is a convenient idealisation of a body with mass but no size.
The laws of motion are simpler for particles than for extended bodies.
If a particle moves along the x-axis, its displacement x(t) from the origin is a function
of time t. We dene the instantaneous velocity v = x and acceleration a = v = x. If a(t)
is known, we can nd v(t) and x(t) by integration:
v(t) =
_
a(t)dt, x(t) =
_
v(t)dt.
Example. If acceleration a = const, we have
v(t) = v
0
+ at, x(t) = x
0
+ v
0
t +
1
2
at
2
,
where the initial conditions are v(0) = v
0
and x(0) = x
0
.
Example. The acceleration of an aircraft during take-o is a = 3 m/s
2
. If the initial
speed v
0
= 0 and the take-o speed is v
1
= 60 m/s, then the take-o time is t
1
= v
1
/a = 20
s, and the required runway length is x(t
1
) x(0) = at
2
1
/2 = v
2
1
/(2a) = 600 m.
The gravitational attraction of the Earth accelerates all falling bodies downwards. The
gravitational acceleration g does not depend on the mass of the body. Near the surface
of the Earth,
z = g, g 9.8 m/s
2
,
33
where we use z to denote the height above ground and neglect air resistance.
Example. Assuming z(0) = h and v(0) = 0, we integrate z = g and get z = gt
and z(t) = h gt
2
/2. The time of the fall t
f
is dened by z(t
f
) = 0. Hence we nd
t
f
=
_
2h/g if air resistance is neglected. For instance, the height of the observation deck
of Auckland Sky Tower is h = 192 m. Consequently the SkyJump duration should be
t
f
=
_
2h/g 6.3 s. The actual duration would be about 11 seconds because you would
be attached to a wire that slows you down.
Example. You drop a stone down a well of unknown depth h and hear the splash T = 4 s
later. What is h, if the speed of sound is c
s
340 m/s and air resistance can be neglected?
The time for the stone to fall down is
_
2h/g, and the time for the sound to travel up is
h/c
s
. It follows that
T =
_
2h/g + h/c
s
.
This is a quadratic equation for

h, which we solve to obtain

h =

_
2/g
_
2/g + 4T/c
s
2/c
s
.
Remembering that

h > 0, we choose the positive root and nd


h =
_
_

_
2/g +
_
2/g + 4T/c
s
2/c
s
_
_
2
72 m.
When gT/c
s
1 we have approximately

2
g
+
4T
c
s
=

2
g

1 +
2gT
c
s

2
g
_
1 +
gT
c
s
_
(remember that

1 +x 1 + x/2 for small x). Therefore, in the limit c


s
, the
formula for h reduces to h = gT
2
/2, as it should.
In three dimensions, the position r of a particle is a vector. We specify a vector by
its components. In Cartesian coordinates we write
r = xi + yj + zk,
where i, j, k are unit vectors, pointing along the three perpendicular axes. We can also
write the same vector simply as r = (x, y, z).
The velocity v(t) of a particle is the time derivative of its position r(t), v = dr/dt.
Similarly the acceleration a(t) is the time derivative of the velocity, a = dv/dt. To
dierentiate a vector, we dierentiate its components. The unit vectors i, j, k do not
depend on time, so their time derivatives vanish. This enables us to write
v = r = ( x, y, z),
34
a = v = r = ( x, y, z).
Example. If r = (1 + t, t
2
, 1 t
3
), we calculate v = r = (1, 2t, 3t
2
) and a = v =
(0, 2, 6t).
Example. The cycloid is the curve traced out by a point on the rim of a wheel when
it rolls along a straight line. If R is the radius of the wheel, and is the angle through
which its radius vector rotates, then the position of a point on the circumference is
r = R( sin , 1 cos , 0).
This is a parametric representation of the cycloid. If the rolling speed R

= const, we
can introduce =

= const and calculate
v = r = R(1 cos t, sin t, 0),
a = v =
2
R(sin t, cos t, 0).
We observe that v = 0 when t = 2k, k = 0, 1, 2, ...: the point where the wheel
touches the ground has a zero instantaneous velocity.
3.2 Dynamics and gravity
Isaac Newtons second law of motion (I. Newton, 1687) states that the acceleration a of
a body of mass m is equal to F/m, where F is the total force acting on the body. We
can write
ma = F.
For motion in a straight line, we have a = x, and so m x = F, where in general the
force F depends on x, x, and t. We can integrate the resulting second-order dierential
equation without much trouble only in a few simple cases.
Example. The slowing-down of a bus is described by m x = F
0
where F
0
= const. If
v(0) = v
0
and x(0) = x
0
, then v(t) = x = v
0
F
0
t/m. Hence the stopping time, dened
by v(t
s
) = 0, is given by t
s
= mv
0
/F
0
. To determine the stopping distance x(t
s
) x(0),
we integrate one more time:
x(t
s
) x
0
= v
0
t
s

F
0
2m
t
2
s
= v
0
mv
0
F
0

F
0
2m
_
mv
0
F
0
_
2
=
mv
2
0
2F
0
.
In three dimensions, a = r and Newtons law of motion for a particle, m r = F, is
equivalent to a system of three second-order dierential equations. To describe the motion
of N particles, we would need to solve a system of 3N equations. Numerical methods are
typically necessary for solving such systems, although sometimes we are able to nd an
exact solution.
35
Example. The motion of a charged particle, say a proton of mass m and electric charge
e > 0, in an electric eld E is described by m r = F, where F = eE. If the electric eld
E = (y, x, 0) where is a positive constant, then we have to solve the system
m x = ey, m y = ex, m z = 0.
Suppose initially r(0) = (x
0
, 0, z
0
) and r(0) = (0, 0, 0). Then the equation m z = 0 is
easily integrated to give z = z
0
. We dene =
_
e/m and rewrite the remaining two
equations as
x =
2
y, y =
2
x.
The symmetry of the system suggests that we introduce new variables u
1
= x + y and
u
2
= x y. On taking the sum and the dierence of the two equations, we get
u
1
=
2
u
1
, u
2
=
2
u
2
.
The initial conditions are u
1
(0) = u
2
(0) = x
0
and u
1
(0) = u
2
(0) = 0. On solving the
resulting uncoupled equations and using the initial conditions, we nd
u
1
(t) = x
0
cosh(t), u
2
(t) = x
0
cos(t).
Finally,
x(t) =
1
2
[u
1
(t) + u
2
(t)] =
x
0
2
[cosh(t) + cos(t)],
y(t) =
1
2
[u
1
(t) u
2
(t)] =
x
0
2
[cosh(t) cos(t)].
A body of mass m near the surface of the Earth experiences the gravitational force
F = (0, 0, mg), where the gravitational acceleration g = 9.8 m/s
2
can be assumed to
be constant in most applications. If no other forces are present, a = F/m reduces to the
simple kinematic equations x = 0, y = 0, z = g. In practice, neglecting air resistance
can lead to huge errors (unless you live on the Moon).
Example. Taking air drag into account, the equation of motion for a falling body is
m z = mg + F
r
,
where F
r
is the resistive force of the air. It often turns out that F
r
= bv
2
is a good
approximation to the air drag, where the speed v = z and b = const. Consequently, we
have a separable dierential equation:
v =
b
m
v
2
g
or
dt =
g
r
2
v
2
g
2
dv,
36
where r
2
= bg/m is introduced for convenience. On integrating using partial fractions,
we get
t =
_
g dv
r
2
v
2
g
2
=
1
2
_
_
1
rv g

1
rv + g
_
dv =
1
2r
ln

rv g
rv + g

+ C.
Suppose the initial condition is v(0) = 0. Then the integration constant C = 0 and
v(t) =
g
r
tanh(rt).
We see that v g/r =
_
mg/b as t . This limiting value of v is called the
terminal speed. For instance, m 70 kg and b 700 kg/m for a falling parachutist,
so the terminal speed is 1 m/s. Finally, replacing v by dz/dt and performing another
integration, we nd the displacement
z z
0
=
_
t
0
v(t)dt =
g
r
_
t
0
tanh(rt)dt =
g
r
2
ln [cosh(rt)] .
Although the gravitational acceleration can be considered constant near the surface
of the Earth, this is not the case when the height above the surface is large enough.
According to Newtons law of gravitation, the force of gravity is inversely proportional to
the square of the distance to the Earths centre:
F = G
mM
E
r
2
.
Here the minus sign indicates attraction, G = 6.6710
11
N m
2
kg
2
is a proportionality
constant, M
E
is the mass of the Earth, m is the mass of a falling body, and r is its distance
from the centre of the Earth. The dimensional units of G mean that when we measure
m and M
E
in kilograms and r in metres, the formula gives us F measured in the units of
force called newtons. Now, as long as the height z above the surface of the Earth is small
compared with the radius of the Earth R
E
, we can approximate r = R
E
+z R
E
, which
leads to
z =
F
m
= G
M
E
(R
E
+ z)
2
G
M
E
R
2
E
.
Numerically, we have R
E
= 6.38 10
6
m and M
E
= 5.98 10
24
kg. Consequently,
g = G
M
E
R
2
E
= 6.67 10
11
5.98 10
24
(6.38 10
6
)
2
m
s
2
9.8
m
s
2
.
We see that g is independent of the mass m of the body, and the gravitational acceleration
can indeed be assumed constant as long as z R
E
.
So far we only discussed situations in which a force acts on a body of constant mass.
If a force F is acting on a body of variable mass m moving with velocity v, Newtons law
of motion asserts that the force F is equal to the rate of change of momentum mv:
F =
d(mv)
dt
.
37
This equation simplies to m r = F when m = const. For motion in one dimension, we
have F = d(mv)/dt.
Example. We have F = 0 for a rocket travelling in interplanetary space. If m(t) and
v(t) are the mass and speed of the rocket, dm < 0 and dv > 0 are their dierentials, and
u = const is the speed of the exhaust gas, expelled by the rocket, then the dierential of
the rocket momentum mdv + vdm must be equal to the dierential of the exhaust gas
momentum (dm)(uv). Here we have to use (uv) rather than u because u is dened
as the speed relative to the moving rocket. Thus we have
mdv + vdm = udm+ vdm
or
mdv + udm = 0.
This is a separable rst-order equation for the speed v as a function of the rocket mass
m. We have
_
v(t)
v(0)
dv = u
_
m(t)
m(0)
dm
m
,
and so
v(t) v(0) = u ln
_
m(0)
m(t)
_
.
This is the so-called rocket equation (K. Tsiolkovsky, 1903). It shows that the speed of
the rocket can be very high if m(0)/m(t) is made large enough.
3.3 Work and energy
For a freely falling body, we integrated z = g and obtained v(t) = z = gt and
z(t) = h gt
2
/2. On eliminating t, we get z = h v
2
/(2g), which yields
1
2
mv
2
+ mgz = mgh,
where the right-hand side does not depend on time. We see that the gain in the kinetic
energy mv
2
/2 is equal to the work done by the gravitational force mg(h z). This result
illustrates the principle of conservation of energy.
More generally, if we have m x = F in one dimension and F = F(x), we can use the
chain rule of dierentiation to write
d
2
x
dt
2
=
dv
dt
=
dv
dx
dx
dt
=
dv
dx
v =
d
dx
_
1
2
v
2
_
.
The equation of motion becomes
m
d
dx
_
1
2
v
2
_
= F(x).
38
On integrating we see that the change in kinetic energy is equal to the work done by the
position-dependent force F(x):
m
2
v
2

m
2
v
2
0
=
_
x
x
0
F(s)ds = U(x),
where the potential energy U(x) =
_
x
x
0
F(s)ds is introduced. We see that the total
particle energy
E =
m
2
v
2
+ U(x)
is a constant of motion because
dE
dt
=
d
dt
_
m
2
v
2
+ U(x)
_
=
d
dt
_
m
2
v
2
0
_
= 0.
The result is valid for any form of F(x).
Example. The motion of a projectile moving radially away from the Earth is described
by r = v and
m v = G
mM
E
r
2
or
v = g
R
2
E
r
2
,
where r is the distance from the centre of the Earth and g = GM
E
/R
2
E
. When r signif-
icantly exceeds the Earths radius R
E
, we can no longer assume that the gravitational
acceleration is constant. To integrate the equation of motion, we rewrite v as
dv
dt
=
dv
dr
dr
dt
=
d
dr
_
1
2
v
2
_
.
On integrating
d
dr
_
1
2
v
2
_
= g
R
2
E
r
2
we obtain for v = v(r):
1
2
_
v
2
v
2
0
_
= gR
2
E
_
1
r

1
R
E
_
,
where we used the initial condition v(R
E
) = v
0
. What is the minimal speed required to
escape Earths gravity? If v v
1
in the limit r , we have v
2
0
= v
2
1
+ 2gR
E
. We
obtain the minimal v
0
by setting v
1
= 0. Hence the required escape speed is
v
0
=
_
2gR
E
11.2 km/s,
where we used g = 9.8 m/s
2
and R
E
= 6.38 10
6
m.
Example. More advanced methods are needed to derive constants of motion for motion
in three dimensions. For instance, it turns out that the energy
E =
m
2
_
x
2
+ y
2
+ z
2
_
exy
39
is a constant of motion for a charged particle of mass m and electric charge e moving in
the electric eld E = (y, x, 0). We can use Newtons equation m r = F with F = eE to
check that

E = m( x x + y y + z z) e( xy + x y)
= (m x ey) x + (m y ex) y + z z
= 0 x + 0 y + 0 z = 0.
3.4 Planetary orbits and Keplers laws
Although it is very natural to think that the Earth is xed at the centre of the Universe
and everything else moves around it, in reality the Earth and other planets move around
the Sun. Johannes Kepler (1609, 1619) analysed a large collection of data on the position
of the planets at various times and established three empirical laws of planetary motion.
1. The orbit of every planet is an ellipse with the Sun in one focus.
2. The line segment from the Sun to a planet sweeps out equal areas in equal times.
3. The square of the orbital period of a planet is proportional to the cube of the
semimajor axis of the ellipse.
Kepler did not have a theory for planetary motion. In a classical example of math-
ematical modelling, Isaac Newton (1687) created such a theory. He discovered how to
derive the inverse square law of gravitational attraction from Keplers laws, and also how
to derive Keplers laws from the inverse square law. We are going to derive Newtons
inverse square law of gravitational attraction from Keplers laws of planetary motion.
The ratio of the mass of a planet to that of the Sun is very smallabout 10
3
for
Jupiter and even smaller for other planets. This is why we can neglect the motion of the
Sun and assume that planets move around a xed point coinciding with the centre of the
Sun. We would need to relax this assumption for binary stellar systems, for instance,
in which two stars of comparable masses orbit around their common centre of mass.
We also assume that the Sun and planets are particles, that is, points at which mass is
concentrated. Newton in fact showed that the Sun and planets behave like particles under
the inverse square law of attraction.
We begin by rewriting Keplers laws mathematically. It is convenient to describe the
motion of a planet around the Sun using polar coordinates. We place the Sun at the
origin of the coordinate system and express the position vector r of a moving particle
(planet) in the form
r = xi + yj = r cos i + r sin j.
Keplers rst law states that the orbit of a planet is an ellipse. The formula for an
ellipse in polar coordinates is
r =
p
1 +e cos
, p > 0, 0 < e < 1.
40
The parameter e is called the eccentricity (not to be confused with the base of natural
logarithms). When e = 0, r = p describes a circle. To show that the formula describes an
ellipse when e > 0, we write the formula as r = p ex and square both sides. We have
r
2
= x
2
+ y
2
= p
2
2epx + e
2
x
2
or
(1 e
2
)x
2
+ y
2
= p
2
2epx,
and so
x
2
+
y
2
(1 e
2
)
= pa 2eax,
where we introduced the length a = p/(1 e
2
) (not to be confused with the acceleration
a). Now we have
x
2
+ 2eax +
a
p
y
2
= pa,
or
(x + ea)
2
+
a
p
y
2
= pa + e
2
a
2
= a
2
,
and nally
(x + ea)
2
a
2
+
y
2
ap
= 1.
This is indeed the usual formula for an ellipse with the semimajor axis a = p/(1e
2
) and
the semiminor axis b =

ap = a

1 e
2
.
If A = A(t) is the area swept out by the position vector r, so that the dierential
dA =
1
2
r
2
d, then Keplers second law states that the rate of change
dA
dt
=
1
2
r
2
d
dt
= const,
and so
r
2

= h = const.
If T is the orbital period of a planet, Keplers third law states that
T
2
a
3
= const,
where the constant on the right-hand side is the same for any planet.
Example. The orbital period of Pluto is T
P
= 248 years. The Earths orbital period,
of course, is T
E
= 1 year. Keplers third law gives the ratio of Plutos semimajor axis to
that of the Earth: a
P
/a
E
= (T
P
/T
E
)
2/3
= 248
2/3
39.5.
To determine the velocity and acceleration of a particle in polar coordinates, we write
r = r cos i + r sin j = r r,
41
where
r = cos i + sin j
is a unit vector in the direction of r. The corresponding unit vector

, perpendicular to
r in the direction of increasing , is given by

= sin i + cos j
(we can check that the two unit vectors are perpendicular by calculating the scalar product
r

= 0). When the angle depends on time t, we have


d r
dt
=
d
dt
(cos i + sin j) = (sin i + cos j)
d
dt
,
and so
d r
dt
=

.
Similarly
d

dt
=

r.
We now use the product rule of dierentiation to obtain the particle velocity
v =
dr
dt
=
d(r r)
dt
=
dr
dt
r + r
d r
dt
= r r + r

and acceleration
a =
dv
dt
=
d
dt
_
r r + r

_
= r r + r

+ r

+ r

2
r
= ( r r

2
) r + (r

+ 2 r

.
We observe that a = a
r
r + a

, where
a
r
= r r

2
, a

= r

+ 2 r

are the radial and azimuthal components of acceleration.


We now use Newtons law of motion
F = ma = m(a
r
r + a

)
and determine the gravitational force F from the acceleration a of a planet. The azimuthal
component of acceleration is
a

= r

+ 2 r

=
1
r
d
dt
_
r
2

_
=
1
r
dh
dt
= 0,
42
since h = const by Keplers second law. Hence the force F has no component perpendic-
ular to r:
F

= ma

= 0, F = F r = ma
r
r.
Next we calculate the component of acceleration a
r
in the radial direction. We need
expressions for r and r. We use Keplers rst and second laws to calculate
r =
d
dt
_
p
1 +e cos
_
=
ep sin
(1 +e cos )
2

=
e
p
(r
2

) sin =
eh
p
sin .
On dierentiating again with respect to time, we nd
r =
d
dt
_
eh
p
sin
_
=
eh
p
cos

=
eh
2
pr
2
cos .
Also,
r

2
=
(r
2

)
2
r
3
=
h
2
r
3
.
Consequently,
a
r
= r r

2
=
eh
2
pr
2
cos
h
2
r
3
=
h
2
r
2
_
e cos
p

1
r
_
=
h
2
r
2
_
e cos
p

1 +e cos
p
_
=
h
2
pr
2
,
and so
F = ma
r
=
mh
2
pr
2
.
To complete the derivation, we need to show that h
2
/p is the same for all planets.
From Keplers second law, dt = r
2
d/h, and the orbital period is
T =
_
T
0
dt =
1
h
_
2
0
r
2
d.
On using Keplers rst law to express r in terms of , we get
T =
p
2
h
_
2
0
d
(1 +e cos )
2
=
p
2
h
2
(1 e
2
)
3/2
.
(The trigonometric integral is dicult, but you may want to try and evaluate it using the
substitution u = tan(/2).) Now Keplers third law gives
T
2
a
3
=
p
4
h
2
4
2
(1 e
2
)
3
a
3
=
4
2
p
h
2
= const,
43
where we used a = p/(1 e
2
). We see that h
2
/p is indeed the same for all planets.
Finally, the force of gravitational attraction should be proportional to the masses of both
gravitating bodies, so we set h
2
/p = GmM, where M is the mass of the Sun and G is
a proportionality constant. We arrive at Newtons inverse square law of gravitational
attraction:
F =
GmM
r
2
r.
Newton postulated that the law is universal: it describes the gravitational force not only
between a planet and the Sun, but also between the Earth and the Moon, for instance,
and in fact between any two masses.
In practice, we can often simplify a problem by assuming that an orbit is circular,
which corresponds to e = 0. For example, the eccentricity of the Earths orbit is only
0.017. If e = 0, we have r = const and r = r = 0. Furthermore,

= h/r
2
= const and

= 0. The orbital speed is v = r

= const and the acceleration a


r
= r

2
= v
2
/r.
Newtons equation of motion ma
r
= GmM/r
2
simplies to
v
2
=
GM
r
, v =
2r
T
.
Example. A telecommunications satellite in a geosynchronous orbit appears to stay
above the same spot. For an orbit in the plane of the equator, we have T = 24 hours,
and the distance between the satellite and the centre of the Earth is r = vT/(2), where
v =
_
GM
E
/r and M
E
is the mass of the Earth. On eliminating v, we nd
r =
_
GM
E
T
2
4
2
_
1/3
42, 200 km,
where G = 6.67 10
11
N m
2
kg
2
and M
E
= 5.98 10
24
kg. The radius of the Earth is
R
E
= 6, 371 km, so the satellite is located approximately 35, 800 km above the sea level.
Example. For a low-orbit satellite, we may assume r = R
E
. We have
v
2
=
4
2
R
2
E
T
2
=
GM
E
R
E
,
and so
T = 2

R
3
E
GM
E
84 minutes.
Alternatively, we can express the low-orbit period T in terms of the gravitational accel-
eration g = GM
E
/R
2
E
or in terms of the average density
E
= M
E
/(
4
3
R
3
E
):
T = 2

R
E
g
=

3
G
E
.
Example. Periods of satellites in elliptical orbits are longer than 84 minutes. For in-
stance, the distance r for the rst Earth-orbiting articial satellite (Sputnik 1, launched
44
in 1957) varied between 6,586 km and 7,310 km. Hence the semimajor axis was a =
(6, 586 + 7, 310)/2 = 6, 948 km, and we obtain from Keplers third law that the period
was T = 84 (6, 948/6, 371)
3/2
96 minutes.
Newtons inverse square law of gravitational attraction implies that when two bodies
are close to each other, gravitational forces are greater in the nearer parts of the bodies.
The resulting tidal forces can be large enough to tear the less massive body apart. The
rings of Saturn, for instance, probably had been produced when the tidal forces ripped
apart its ancient moon.
Exercises
1. The motion of a slowing train is described by
m v = v
2
, v(0) = v
0
,
where and are positive constants. Calculate the stopping time of the train t
s
. Deter-
mine the limit of t
s
as v
0
.
2. At what altitude above the surface of a planet is the gravitational acceleration one
half of that at the surface? The radius of the planet is R.
3. Calculate the work required to raise a body of mass m from the surface of the Earth
to a height h above the surface. Assume that the radius of the Earth is R and the
gravitational acceleration near the surface is g. Also calculate the required work in the
limit h .
4. The Earth orbits around the Sun with a period of 1 year. The average SunEarth
distance is about 1.5 10
8
km. Use these facts to determine the mass of the Sun.
Comment: how can we determine the SunEarth distance? If you are curious, read about
it at http://curious.astro.cornell.edu
5. (a) The orbital period of a planet is 27 years. What is the ratio of its semimajor axis
to that of the Earth? (b) Determine the orbital period (in years) of a planet whose mean
distance from the Sun is 25 times that of the Earth.
6. The Vostok-1 spacecraft carried the rst human into outer space in 1961. The height
of the orbit was 181 km in its perigee and 327 km in its apogee. Determine the orbital
period of the spacecraft. You may assume that the period of a low-orbit satellite is 84
minutes, and the average radius of the Earth is 6371 km.
7. If the Earth suddenly stopped orbiting the Sun, it would be pulled in by the gravita-
tional force of the Sun. How long would it take for the Earth to hit the Sun?
8. Suppose you are travelling to the antipodes in a low-orbit satellite. What would be
your travel time? Express your answer in terms of the gravitational acceleration at the
surface, g, and the radius of the Earth, R.
45
9. A spaceship is in a low circular orbit around a newly discovered planet which is made
of pure platinum. The density of platinum is four times greater than the average density
of the Earth. The period of a low-orbit satellite orbiting around the Earth is about 90
minutes. What is the orbital period of the spaceship?
46
Chapter 4
Systems of rst-order dierential
equations
4.1 Homogeneous linear systems with constant coef-
cients
In many applications, we need to determine several functions that are described by a
system of dierential equations. We have already encountered examples of Richardsons
theory of conict and Lanchesters model of combat, which lead to such systems. Another
example is provided by two-species models in biology (deer and the vegetation they con-
sume, foxes and rabbits, sharks and smaller sh). Thus we are often interested in solving
a system of two rst-order dierential equations:
x = f(x, y),
y = g(x, y),
where f and g are some known functions.
We start with homogeneous linear systems with constant coecients:
x = ax + by,
y = cx + dy,
where a, b, c, d are constants. We can always solve this system by eliminating one of
the unknown functions and reducing the system to a single second-order equation with
constant coecients. If c = 0, for example, we can eliminate x by rewriting the second
equation as
x =
1
c
y
d
c
y,
which we dierentiate to nd
x =
1
c
y
d
c
y.
47
On substituting these x and x into the rst equation, we get a homogeneous second-order
equation
y (a + d) y + (ad bc)y = 0.
As usual, we seek solutions of the form y(t) exp(t) and obtain the quadratic equation

2
(a + d) + (ad bc) = 0
with the two roots

1,2
=
1
2
_
(a + d)
_
(a + d)
2
4(ad bc)
_
.
Assuming that
1
=
2
, the general solution for y(t) is
y(t) = c
1
e

1
t
+ c
2
e

2
t
,
where the constants c
1
and c
2
are specied by initial conditions. It follows that the general
solution for x(t) is
x(t) =
1
c
y
d
c
y
= c
1

1
d
c
e

1
t
+ c
2

2
d
c
e

2
t
.
Of course, we would obtain the same general solution of the system if we eliminated y
and solved the resulting second-order equation for x.
Example. In order to solve
x = 3x y, x(0) = 3,
y = x + y, y(0) = 0,
we eliminate y. From the rst equation, we have y = x + 3x, and so y = x + 3 x.
Substitution into the second equation yields
x 4 x + 4x = 0,
which leads to

2
4 + 4 = 0.
Because
1
=
2
= 2, the general solution is
x(t) = (c
1
+ c
2
t)e
2t
and
y(t) = x + 3x = (c
1
c
2
+ c
2
t)e
2t
On using the initial conditions x(0) = c
1
= 3 and y(0) = c
1
c
2
= 0, we obtain c
1
= c
2
= 3,
and thus the unique solution is
x(t) = 3(1 +t)e
2t
, y(t) = 3te
2t
.
48
4.2 Equilibrium and stability: linear systems
For a general system
x = f(x, y),
y = g(x, y),
we dene an equilibrium solution, say x = x
0
, y = y
0
, as a solution that does not vary
in time: x
0
= y
0
= 0. An equilibrium is called stable if an arbitrary small displacement
x(t), y(t) from the equilibrium remains in the vicinity of the equilibrium. Otherwise,
the equilibrium is called unstable. The equilibrium is called asymptotically stable if, in
addition to stability, x(t) x
0
and y(t) y
0
as t . These denitions for a system
are similar to those for a single dierential equation.
It will be seen later that the stability of equilibrium solutions of a nonlinear system
can be determined by studying a related linear system. This is why it is instructive to
begin by analysing linear systems which we can always solve exactly. We observe that
x
0
= y
0
= 0 is an equilibrium solution of the linear system with constant coecients
x = ax + by,
y = cx + dy.
We can express both x(t) and y(t) as linear combinations of exponential solutions exp(
1
t)
and exp(
2
t), where

1,2
=
1
2
_
(a + d)
_
(a + d)
2
4(ad bc)
_
.
Because we can reduce the linear system to a second-order homogeneous equation with
constant coecients, we can obtain a stability criterion for the linear system using the
argument that we used to obtain the stability criterion for a second-order equation. We
require for stability that the real parts of both
1
and
2
be negative. This is equivalent
to

1
+
2
< 0,
1

2
> 0,
or
a + d < 0, ad bc > 0.
The equilibrium x
0
= y
0
= 0 is asymptotically stable if and only if both these conditions
are satised. It is important that we can determine whether or not the equilibrium is
stable without writing down the general solution of the system. Unless otherwise noted,
we do not investigate the more dicult question of neutral stability. Qualitative theory
of ordinary dierential equations is used to describe the behaviour of solutions in the
vicinity of equilibrium points.
Example. For the system
x = 3x y,
y = x + y,
we have a+d = 3+1 = 4 > 0 and adbc = 31(1) 1 = 4 > 0. Because a+d > 0,
the equilibrium x
0
= y
0
= 0 is unstable.
49
4.3 Equilibrium and stability: nonlinear systems
To determine an equilibrium solution x = x
0
, y = y
0
of the system
x = f(x, y),
y = g(x, y),
we solve
f(x
0
, y
0
) = 0,
g(x
0
, y
0
) = 0.
In applications, we are usually interested in stable equilibrium solutions that are not
destroyed by small perturbations.
To determine whether an equilibrium is stable, we need to know the behaviour of a
solution x(t), y(t) that is initially very close to the equilibrium. Therefore, just as for rst-
order and second-order dierential equations, we can use a linearised stability analysis to
investigate whether an equilibrium solution is asymptotically stable. We dene a time-
dependent deviation from the equilibrium
u(t) = x(t) x
0
, u = x,
v(t) = y(t) y
0
, v = y,
and we expand f(x, y) and g(x, y) in Taylor series as functions of two variables:
f(x, y) = f(x
0
, y
0
) +
f
x

x
0
,y
0
(x x
0
) +
f
y

x
0
,y
0
(y y
0
) + . . . ,
g(x, y) = g(x
0
, y
0
) +
g
x

x
0
,y
0
(x x
0
) +
g
y

x
0
,y
0
(y y
0
) + . . . .
Keeping only the linear terms in the expansions and using f(x
0
, y
0
) = 0 and g(x
0
, y
0
) = 0,
we approximate the original nonlinear system by a linear system with constant coecients:
u = au + bv,
v = cu + dv,
where
a =
f
x

x
0
,y
0
, b =
f
y

x
0
,y
0
, c =
g
x

x
0
,y
0
, d =
g
y

x
0
,y
0
.
We have already derived the asymptotic stability criterion for equilibrium solutions of
linear systems:
a + d < 0, ad bc > 0.
Now we can determine whether the solution of a nonlinear system is asymptotically stable
by applying this criterion to a linearised system.
50
We can only use this approach to investigate asymptotic stability. As the following
example shows, the linearised system cannot be used to establish whether an equilibrium
solution is neutrally stable.
Example. The solution x
0
= 0, y
0
= 0 is the only equilibrium of the system
x = y + x(x
2
+ y
2
),
y = x + y(x
2
+ y
2
).
The linearised system
u = v, v = u
is equivalent to
u + u = 0, v + v = 0.
Because its solutions are oscillatory and hence remain small for all t, we might draw the
wrong conclusion that x
0
= 0, y
0
= 0 is a neutrally stable equilibrium of the nonlinear
system as well. However, if we consider the function s(t) = x
2
+ y
2
, we see that
1
2
ds
dt
= x x + y y = x[y + x(x
2
+ y
2
)] + y[x + y(x
2
+ y
2
)] = (x
2
+ y
2
)(x
2
+ y
2
)
or
s = 2s
2
.
On integrating this separable rst-order equation, we nd
s(t) =
s(0)
1 2s(0)t
.
It is clear that s = x
2
+ y
2
as t [2s(0)]
1
, which can only happen if either x
or y or both become innite as t [2s(0)]
1
. Therefore, x
0
= 0, y
0
= 0 is an unstable
equilibrium.
Example. Previously we used the logistic equation

N = a(1 N/K)N, where a and
K are positive constants, to describe the limiting eect of nite resources on the size of
a population. Now consider two species that are competing for the same limited food
supply. Introducing a parameter > 0 that measures the degree of competition between
the two species for the resources, we can write for the two populations N
1
and N
2

N
1
= a
1
_
1
N
1
K
1
N
2
_
N
1
,

N
2
= a
2
_
1
N
2
K
2
N
1
_
N
2
.
For simplicity, we assume K
1
= K
2
= . The resulting nonlinear system

N
1
= f(N
1
, N
2
) = a
1
(1 N
2
) N
1
,
51

N
2
= g(N
1
, N
2
) = a
2
(1 N
1
) N
2
has two equilibrium solutions: N
10
= N
20
= 0 and N
10
= N
20
=
1
.
To investigate the stability of the equilibria, we introduce the perturbations u(t) =
N
1
(t) N
10
, v(t) = N
2
(t) N
20
and linearise the system about an equilibrium. For the
solution N
10
= N
20
= 0, the linearised system is
u = a
1
u,
v = a
2
v,
and we conclude that the equilibrium is unstable.
For the nonzero equilibrium solution N
10
= N
20
=
1
, we calculate the partial
derivatives
a =
f
N
1

eq
= a
1
(1 N
20
) = 0,
b =
f
N
2

eq
= a
1
N
10
() = a
1
,
c =
g
N
1

eq
= a
2
N
20
() = a
2
,
d =
g
N
2

eq
= a
2
(1 N
10
) = 0,
where the subscript eq indicates equilibrium values. The linearised system is
u = 0u a
1
v,
v = a
2
u + 0v.
To apply our stability criterion, we calculate a + d = 0 + 0 = 0 and ad bc = 0
0 (a
1
) (a
2
) = a
1
a
2
. We see that neither a + d < 0 nor ad bc > 0 is satised.
Hence the model predicts that the equilibrium is unstable: two similar species in the same
habitat cannot coexist. The same result is obtained for nite K
1
and K
2
. Therefore we
should expect that the struggle for existence between two similar species nearly always
leads to the complete extinction of one of them. This result is known as the principle of
competitive exclusion in population biology.
Example. Suppose x(t) is the number of males infected with gonorrhoea, y(t) is the
number of infected females, a
1
and a
2
are the rates with which the infectives are cured, c
1
=
const is the number of promiscuous males, and c
2
= const is the number of promiscuous
females. We assume that new male infectives are added at a rate b
1
(c
1
x)y that is
proportional to the number of the male susceptibles c
1
x and to the number of the
female infectives y. Similarly, we assume that new female infectives are added at a rate
b
2
(c
2
y)x that is proportional to the number of the female susceptibles c
2
y and to
52
the number of the male infectives x. Then a model for the spread of gonorrhoea due to
heterosexual contacts is
x = a
1
x + b
1
(c
1
x)y,
y = a
2
y + b
2
(c
2
y)x.
For practical reasons, we are interested in the equilibrium solution x
0
= 0, y
0
= 0. Is
it stable? The linearised system is
x = a
1
x + b
1
c
1
y,
y = b
2
c
2
x a
2
y.
We observe that a+d = (a
1
+a
2
) < 0, and thus the rst part of our stability criterion is
always satised. Now, ad bc = (a
1
)(a
2
) (b
1
c
1
)(b
2
c
2
), and the equilibrium is stable
if ad bc > 0 or
b
1
b
2
c
1
c
2
< a
1
a
2
.
This is when the disease disappears from the population.
Otherwise, if b
1
b
2
c
1
c
2
> a
1
a
2
, it can be checked by a tedious calculation that
x
0
=
b
1
b
2
c
1
c
2
a
1
a
2
a
1
b
2
+ b
1
b
2
c
2
, y
0
=
b
1
b
2
c
1
c
2
a
1
a
2
a
2
b
1
+ b
1
b
2
c
1
is a stable equilibrium. In this case the total number of infective males and females
ultimately levels o. It should not surprise us that public health data indicate that
b
1
b
2
c
1
c
2
> a
1
a
2
.
Example. Consider now two species, one of which feeds on the other. Suppose x(t) is
the prey population, say small sh, and y(t) is the predator population, say sharks. If the
species did not interact, we could write x = rx, where r > 0 implies abundant food supply
for the prey population, say algae for the small sh. By contrast, the predators would die
o without food, and so y = ky. In reality the two species do interactthe small sh are
eaten by the sharksand the number of contacts between them is proportional to both
x and y. Hence we arrive at the following system describing predatorprey interactions:
x = f(x, y) = rx xy,
y = g(x, y) = ky + xy.
These equations are called the LotkaVolterra equations (A. J. Lotka, 1925; V. Volterra,
1926).
We nd equilibrium solutions by solving
f(x
0
, y
0
) = x
0
(r y
0
) = 0,
g(x
0
, y
0
) = (x
0
k)y
0
= 0.
53
The system has two equilibrium solution. The solution x
0
= 0, y
0
= 0 is unstable since
the corresponding linearised system is
x = rx, y = ky,
and so a small nonzero x will deviate strongly from the equilibrium. For the nonzero
equilibrium solution x
0
= k/, y
0
= r/, we calculate the partial derivatives
a =
f
x

eq
= r y
0
= 0,
b =
f
y

eq
= x
0
=
k

,
c =
g
x

eq
= y
0
=
r

,
d =
g
y

eq
= x
0
k = 0,
and we use them to obtain the linearised system
u = 0u + (k/)v,
v = (r/)u + 0v.
To apply our stability criterion, we calculate a + d = 0 + 0 = 0 and ad bc = 0 0
(k/)(r/) = rk > 0. We see that a+d < 0 is not satised, and thus the equilibrium
is not asymptotically stable. More detailed analysis, however, shows that the equilibrium
is neutrally stable. On dierentiating the linearised equations, we obtain
u + kru = 0, v + krv = 0.
Thus the key prediction of the LotkaVolterra model is that the predator and prey popu-
lations should oscillate with period T = 2/

kr. In the real world, of course, interaction


between species are a lot more complicated than in this one-predator-one-prey model.
Example. The Italian biologist Umberto DAncona noticed that the total catch in the
Mediterranean had a larger fraction of sharks during World War I when the level of shing
was greatly reduced. To explain this phenomenon, the mathematician Vito Volterra
incorporated the eect of shing into the LotkaVolterra model. If h is the fraction of
sharks and small sh being caught, we can write
x = rx xy hx,
y = ky + xy hy.
54
It follows that the equilibrium solution is
x
0
=
k + h

, y
0
=
r h

.
Therefore, as long as h < r, harvesting increases the equilibrium prey population and
decreases the equilibrium predator population. (There is no biologically meaningful equi-
librium in the case h > r that corresponds to overshing.) This is why reduced shing
during the war (smaller h) had increased the shark population (larger y
0
) and decreased
the average sh population (smaller x
0
). More recently, an unanticipated eect of DDT
spraying was to destroy not the harmful insects but rather their natural predators.
Exercises
1. Suppose R(t) measures Romeos love for Juliet and J(t) measures Juliets love for
Romeo. Romeos love grows in response to Juliets love for him and vice versa. At the
same time, both lovers try to rein in their feelings for each other. The resulting system
of equations is as follows:

R = R + J,

J = J + R.
Find the general solution of the system. Solve the initial value problem R(0) = 1, J(0) =
0.
2. We wish to model the outbreak of an infectious disease. The number of individuals
susceptible to infection at time t is S(t), and the number of infected individuals is I(t). We
postulate that the population increases at a constant rate in the absence of an epidemic.
We also postulate that the infection rate is proportional to the product of S and I and that
the death rate is proportional to I. Consider the resulting system of coupled dierential
equations for S(t) and I(t):

S = rSI + ,

I = rSI I.
Here r, and are positive constants. Derive an equilibrium solution of the system,
S = S
0
and I = I
0
, and determine its stability.
3. We wish to model the size of an insect population. The population at time t is P(t),
and the total food supply is F(t). We postulate that the birth rate is proportional to the
product of the population size and the food supply and that the death rate is proportional
to the population size. We also postulate that F(t) is described by a linear rst-order
dierential equation in the absence of the insects and that the food consumption rate is
proportional to the population size. Consider the resulting system of coupled dierential
equations for P(t) and F(t) in appropriate units:

P = FP kP,
55

F = c F P.
Here k and c are positive constants, and we assume c > k. Derive an equilibrium solution
of the system, P = P
0
> 0 and F = F
0
, and determine its stability.
4. We wish to model the size of a sh population in a lake. The sh population at time t
is F(t), and the number of shermen is M(t). We postulate that sh grow logistically in
the absence of shing, and that the presence of shermen decreases the growth rate by an
amount proportional to the product of the sh and shermen numbers. We also postulate
that shermen are attracted to the lake at a rate proportional to the amount of sh in
the lake, and that shermen are discouraged from the lake at a rate proportional to the
number of shermen already there. Consider the resulting system of coupled dierential
equations for F(t) and M(t) in appropriate units:

F = aF(1 F) FM,

M = bF M.
Here a and b are positive constants. Calculate the equilibrium solutions of the system
and determine their stability.
56
Chapter 5
Dierence equations
5.1 Geometric growth
In many applied problems, we are interested in variables that are only known at certain
time intervals. For example, economic quantities (income, savings, gross domestic prod-
uct, unemployment) can be reported weekly, monthly or annually. Learning results are
evaluated by tests and exams. Population sizes of annual plants vary from year to year.
These problems lead to mathematical models that involve dierence equations.
The dierence equation for geometric growth is
x
k+1
= ax
k
, a = const,
where x
k
is some variable at time step k. Given an initial value x
0
, we can determine
x
1
= ax
0
, x
2
= ax
1
= a
2
x
0
, and in general
x
k
= x
0
a
k
.
Example. If a sum of money accumulates at compound interest, S
0
is the initial deposit,
and S
k
is the sum on deposit after k compounding intervals, then S
k+1
= S
k
+ iS
k
=
(1 +i)S
k
, where i is the interest rate per interval. Hence S
k
= S
0
(1 +i)
k
.
Example. In a model for economic growth, we consider annual national income Y
k
for
year k. The total income is divided into consumer expenditures C
k
and private investments
I
k
,
Y
k
= C
k
+ I
k
.
We assume that consumption is proportional to the available income,
C
k
= mY
k
, 0 < m < 1,
and also that income growth is proportional to the invested amount,
Y
k+1
Y
k
= rI
k
, r > 0.
57
We have
Y
k+1
= Y
k
+ rI
k
= Y
k
+ r(Y
k
C
k
)
= Y
k
+ r(Y
k
mY
k
)
= (1 +r rm)Y
k
or
Y
k
= Y
0
(1 +r rm)
k
.
The model neglects many important factors, such as unemployment, so its prediction of
a robust economic growth is not very realistic.
5.2 The equation x
k+1
= ax
k
+ b
The dierence equation
x
k+1
= ax
k
+ b, a = const, b = const
is a linear rst-order equation with constant coecients. The equation is rst-order
because it only involves x
k
and x
k+1
. If a = 1, its solution is x
k
= x
0
+ kb.
Example. If a sum of money earns simple interest at the rate r, S
0
is the initial deposit,
and S
k
is the sum on deposit after k years, then S
k+1
= S
k
+ rS
0
and S
k
= S
0
(1 +kr).
If a = 1, we seek a solution of the form x
k
= C
1
+ C
2

k
, where C
1
, C
2
, and are
constants to be determined. On substituting into the equation, we get
C
1
+ C
2

k+1
= aC
1
+ aC
2

k
+ b
or
(C
1
aC
1
b) + C
2

k
( a) = 0,
which we satisfy by choosing
C
1
=
b
1 a
, = a.
We use the initial condition x
0
= C
1
+ C
2
to specify the constant C
2
= x
0
C
1
=
x
0
b/(1 a). The resulting solution is
x
k
=
b
1 a
+
_
x
0

b
1 a
_
a
k
or
x
k
= x
0
a
k
+ b
1 a
k
1 a
.
If b = 0, we recover the formula for geometric growth.
58
Behaviour of solutions to rst-order dierence equations is very dierent for dierent
numerical values of a and b. A solution x
k
can monotonically decrease or increase (with
or without a nite limit as k ), oscillate with a constant amplitude, or oscillate with
increasing or decreasing amplitude.
Example. The solution of the dierence equation
x
k+1
= 2x
k
+ 1, x
0
= 5
is x
k
= 6 2
k
1. We observe that x
k
as k .
Example. The solution of the equation
x
k+1
=
1
2
x
k
+ 2, x
0
= 3
is x
k
= 4 (1/2)
k
. We observe that x
k
4 as k .
Example. The solution of the equation
x
k+1
= x
k
+ 1, x
0
= 1
is x
k
= [1 + (1)
k
]/2. We observe that x
k
oscillates with a constant amplitude.
Example. The quantity supplied S
k+1
of some commodity (say, crop harvested and
brought to market) is an increasing function of its price P
k
in the previous year. The
quantity demanded D
k+1
of the same commodity is a decreasing function of the current
price. Suppose that
S
k+1
= aP
k
, D
k+1
= b cP
k+1
,
where a, b, c are positive constants. Market equilibrium is reached when supply equals
demand, that is S
k+1
= D
k+1
for any k. Then we have aP
k
= b cP
k+1
or
P
k+1
=
b
c

a
c
P
k
,
and so
P
k
=
b
a + c
+
_
P
0

b
a + c
_
_

a
c
_
k
.
We note that price equilibrium can be achieved only if a/c < 1, in which case P
k

b/(a + c) as k . Otherwise, the model predicts that the commodity price will
oscillate with an increasing amplitude.
5.3 Fibonacci numbers
The Fibonacci numbers F
k
are dened by
F
k+2
= F
k+1
+ F
k
, F
0
= F
1
= 1,
59
so that F
k
= 1, 1, 2, 3, 5, 8, 13, . . .. This equation is an example of a linear second-order
dierence equation with constant coecients. In 1202, the Italian merchant and math-
ematician Leonardo Fibonacci introduced the sequence F
k
to describe the growth of a
(biologically unrealistic) rabbit population: F
k
is the population after k months if rabbits
start to breed after two months and each month each pair begets a new pair. We may not
care about the peculiar habits of incestuous rabbits, but the Fibonacci numbers appear
in more plausible contexts. For instance, F
k
is the number of branches on a tree if a new
shoot has to grow two months before it branches.
Just as we solve homogeneous linear dierential equations with constant coecients by
seeking solutions exp(t), we can solve linear dierence equations by seeking solutions

k
. On substituting F
k
= C
k
into the Fibonacci equation, we obtain the quadratic
equation

2
1 = 0
with the roots

1
=
1
2
(1 +

5),
2
=
1
2
(1

5).
Therefore the general solution of the Fibonacci equation is
F
k
= C
1

k
1
+ C
2

k
2
.
We use the initial conditions F
0
= C
1
+C
2
= 1 and F
1
= C
1

1
+C
2

2
= 1 to specify the
constants
C
1
=
1
2

2
=
1 +

5
2

5
=
1

1
,
C
2
= 1 C
1
=

5 1
2

5
=
1

2
.
Therefore a formula for the Fibonacci sequence is
F
k
=
1

5
_
1 +

5
2
_
k+1

5
_
1

5
2
_
k+1
.
It is perhaps not obvious that this formula gives a natural number for any k.
5.4 Nonlinear dierence equations
There is no general method for solving nonlinear dierence equations, but sometimes we
are able to nd a solution by guessing a change of variables that reduces a nonlinear
equation to a simpler equation.
Example. R. J. H. Beverton and S. J. Holt (1957) suggested a nonlinear model for the
size of a sh population n
k
in year k:
n
k+1
=
Rn
k
1 +n
k
/M
,
60
where R > 1 and M > 0 are constants, and (1 + n
k
/M)
1
is the probability of survival
till next year. The BevertonHolt equation is a discrete analogue of the logistic growth
equation. Substitution x
k
= 1/n
k
leads to a linear rst-order equation
x
k+1
=
1
R
x
k
+
1
RM
with the solution
x
k
= x
0
R
k
+
1 R
k
(R 1)M
or
n
k
=
1
x
k
=
Qn
0
n
0
+ (Qn
0
)R
k
,
where Q = (R1)M. Because n
k
Q as k , we see that Q is the carrying capacity
in the BevertonHolt model.
Example. The nonlinear equation
x
k+1
= 2x
k
(1 x
k
)
can be rewritten as
1 2x
k+1
= (1 2x
k
)
2
.
It follows that
1 2x
k
= (1 2x
0
)
2k
or
x
k
=
1
2
(1 2x
0
)
2k
.
We observe that x
k

1
2
as k , if |1 2x
0
| < 1 or 0 < x
0
< 1.
Example. Consider the nonlinear equation
x
k+1
= 4x
k
(1 x
k
).
Using the substitution x
k
= sin
2

k
, we get
sin
2

k+1
= 4 sin
2

k
cos
2

k
or
sin
k+1
= sin(2
k
),
and so, as long as /2 <
k
< /2, we have
k
=
0
2
k
and
x
k
= sin
2
(
0
2
k
).
We observe that a very small change in
0
would lead to a very dierent solution after
only a few steps because of the rapidly increasing factor 2
k
.
61
When we cannot solve a nonlinear dierence equation analytically, we can try to learn
about the properties of its solutions by studying equilibrium points and their stability.
For a dierence equation
x
k+1
= f(x
k
),
where f is a known function, we dene an equilibrium point as a constant x
e
such that
x
e
= f(x
e
).
A small deviation from the equilibrium is described by
x
k+1
x
e
= f(x
k
) f(x
e
) f

(x
e
)(x
k
x
e
).
Here f

(x
e
) is the derivative of f(x
k
) with respect to x
k
, evaluated at x
k
= x
e
. We see
that x
k
x
e
as k , if |f

(x
e
)| < 1. Hence we can use the criterion |f

(x
e
)| < 1 to
determine whether an equilibrium is stable.
Example. The nonlinear dierence equation
x
k+1
= f(x
k
) = rx
k
(1 x
k
), 0 < r 4
is called the logistic map. This equation can be solved if r = 2 or r = 4 (see the previous
two examples). In general, the logistic map has two equilibrium points: x
e1
= 0 and
x
e2
= 11/r. We calculate f

(x
k
) = r2rx
k
, which yields f

(x
e1
) = r and f

(x
e2
) = 2r.
Hence x
e1
= 0 is stable if 0 < r < 1, and x
e2
= 1 1/r is stable if 1 < r < 3. We note
that x
e2
=
1
2
if r = 2, which agrees with the exact solution of the equation.
Suppose we x the parameter r, choose some x
0
(0 < x
0
< 1) and use the logistic
map x
k+1
= rx
k
(1 x
k
) to calculate x
1
, x
2
, . . . . We expect that x
k
should tend to
a stable equilibrium in the limit k . Using the previous example, we see that
x
k
x
e1
= 0 if 0 < r < 1, and x
k
x
e2
= 1 1/r if 1 < r < 3. What happens for
larger values of r when there are no stable equilibrium solutions? This is an interesting
question that motivated a large number of both numerical and analytical studies. It
turns out that if 3 < r < 3.449 . . ., x
k
oscillates with period 2, that is x
k+2
= x
k
.
If 3.449 . . . < r < 3.544 . . ., x
k
oscillates with period 4, that is x
k+4
= x
k
. Solutions
with longer and longer periods appear for larger r, until the system becomes chaotic for
r > 3.569 . . .. Chaotic means that the solution x
k
never repeats itself and is extremely
sensitive to the initial condition x
0
. Thus even simple nonlinear equations can have
solutions with very complicated behaviour.
Exercises
1. Suppose that a species of animal only breeds during the spring and all adults die
before the next breeding season. Every female produces on average R female ospring
that survive to breed in the next year. Find the female population as a function of time.
62
2. A model for the price of bonds leads to the dierence equation
p
t
=
_
1
E
D
_
p
t1
+
E
D
P, t = 1, 2, 3, . . . ,
where p
t
denotes the bond price at time t, p
0
is known, and E, D, and P are constants.
Solve this equation and thus show that p
t
converges to P if 0 < E/D < 2.
3. A sum P is deposited in a bank at equal intervals. The interest rate per interval is i.
If S
k
is the sum on deposit after k intervals, show that
S
k+1
= (1 +i)S
k
+ P.
Find the solution of this dierence equation if S
0
= 0.
4. Suppose that B dollars are borrowed from a bank at a xed annual interest rate R%.
The loan is to be repaid in n years in equal monthly instalments of P dollars. Find P.
5. The Fibonacci numbers F
k
are dened by F
k+2
= F
k+1
+ F
k
. Calculate lim
k
F
k+1
F
k
.
6. Consider the dierence equation
x
k+2
3x
k+1
+ 2x
k
= 0.
Show that the general solution to this equation is x
k
= C
1
+ C
2
2
k
. Find the particular
solution such that x
0
= 1, x
1
= 2.
7. A model for the number of red blood cells circulating the blood is given by
R
k+1
= (1 f)R
k
+ M
k
,
M
k+1
= fR
k
,
where R
k
is the number of cells in the circulatory system on day k, M
k
is the number of
cells produced by marrow on day k, and f is a fraction of cells removed by spleen daily.
Derive a single second-order equation for R
k
and nd its general solution.
8. Solve the nonlinear dierence equation x
k+1
= x
k
/(1 +x
k
).
9. Find the nonzero equilibrium point of the dierence equation x
k+1
= (x
k
)
2
and deter-
mine whether it is stable.
10. A model for the size of sh populations leads to the nonlinear dierence equation
N
k+1
= N
k
exp(N
k
),
where and are positive constants. Find the nonzero equilibrium solution of this
equation and determine when it is stable.
11. The nonlinear dierence equation
x
k+1
= f(x
k
) = rx
k
(1 x
k
)
is known to have stable period-2 solutions, x
k+2
= x
k
, when the parameter r lies in the
range r
1
< r < r
2
. Calculate the values of r
1
and r
2
by considering the equation
x
k+2
= f
2
(x
k
) = f(f(x
k
)).
63

You might also like