C H A P T E R 6

Modeling with Discrete

Dynamical Systems

6.1 INTRODUCTION
One of the most exciting areas of modeling concerns predicting temporal evolution.
The main question that is posed in this setting is how do variables of interest change
over time? This type of problem is everywhere to be found, for example in areas as
diverse as science, engineering and finance. Prediction means that given the values
of the variables at a certain instant of time we can predict, i.e. compute their values
at any future time. A system of equations that allows such a prediction is called a
Dynamical System.
In this chapter we consider discrete dynamical systems. The mathematical
assumption is that the time variable n is incremented discretely and corresponds
to the integers {0, 1, 2, 3, 4, . . . }. The value of a variable x of interest is then a
sequence {x0 , x1 , x2 , x3 , x4 , . . . }. Now the problem of modeling is to determine an
equation of the form
xn+1 = xn + ∆xn
and this is done by estimating how the variable xn changes as n is incremented
from time n to time n + 1.
We develop this topic along the following four complementary lines:

• numerical solutions,
• analytical solutions,
• qualitative behavior,
• modeling techniques.

As the terminolgy suggests, numerical approaches to difference equations will in-

volve direct computation of these sequences via computer. In contrast, analytical
solution methods seek closed form solutions; these are available only in limited
circumstances.
Qualitative approaches are analytical as well as numerical approaches to de-
termine the qualitative behaviour of the solutions in the long run. The questions
addressed are: do the solutions go off to infinity, do they approach a finite value,
will they oscillate or behave more complicated? Another question of interest is
the sensitivity of solutions to variation of parameters. A change in the qualitative
behaviour when a parameter is varied is called a bifurcation.
The topic of modeling will treat empirical and qualitative approaches for
constructing difference equations. We will consider the development of models

106
 Section 6.1 Introduction 107

2.5 2.5
x0=2.5
x0=1 x =0.5*x +1
2 n+1 n

2
1.5

n
x
xn 1
1.5

0.5
xn+1=0.5*xn
x =2.5
0
x =1
0
0 1
0 5 n 10 15 0 5 n 10 15

(a) (b)

8000 2.5
x =0.2 x =2.5
0
0
x0=−0.2 2 x =−1
0
4000
1.5

1
n

xn
x

0
0.5

−4000 0
x =2*x
n+1 n
−0.5 x =−0.5*x +1
n+1 n
−8000 −1
0 5 n 10 15 0 5 n 10 15

(c) (d)

FIGURE 6.1: Comparison of the numerical solutions for some simple difference equations.

based on the qualitative approaches presented in Chapter 2 as well as the more

quantitative data fitting approaches of Chapter 5.
A simple but nevertheless important difference equation is the equation

xn+1 = axn + b. (6.1)

If an initial value x0 is fixed the solution is determined for all n,

x1 = ax0 + b, x2 = ax1 + b, x3 = ax2 + b, . . . .

Numerically simulated solutions of (6.1) for various values of the parameters a and
b are shown in Figure 6.1. In Figure 6.1 (a) we see that the solutions decay to zero
while in Figure 6.1 (b) they tend to the value 2. In Figure 6.1 (c) the initial values
are close to zero. Both solutions remain close to zero for a while, but eventually
they split apart and tend to ±∞. In Figure 6.1 (d) the solutions tend to x ≈ 0.7.
Here the solutions alternate between values above and below 0.7 when approaching
108 Chapter 6 Modeling with Discrete Dynamical Systems

this value. Thus, a noticeable feature for all of these solutions is the long term
behavior. Qualitatively we say the solution either blows up or approaches a finite
limiting value.

EXAMPLE 6.1 Discrete Compound of Interest

Interest rates for loans or saving accounts are normally fixed on an annual basis,
however the compounding scheme typically applies the interest charges monthly.
Suppose you purchase something for a certain amount of $a0 and charge it to your
credit card that carries an annual interest rate of r%. Let an be the accumulated
debt after n months. In Section 6.2.2 we will see that an satisfies the difference
equation
r
an+1 = (1 + )an − p, (6.2)
1200
where p is your monthly payment. Equation (6.2) has the form of Equation (6.1).
By solving this equation you can answer questions such as: when is a loan a0 paid
off given a certain monthly payment p, or what should the monthly payment be in
order that the loan is paid off after a prescribed amount of time?

Equation (6.1) is called a linear first order difference equation. It is linear

because the right hand side is a linear function of xn . It is of first order because
only one time step is involved. The simplest nonlinear first order difference equation
is
xn+1 = axn + bx2n . (6.3)
In Figure 6.2 numerical solutions of (6.3) are shown for b = −1 and two different
values of a. In Figure 6.2 (a) we see approach to a limiting value as in Figure 6.1
(d). In contrast in Figure 6.2 (b) the solution eventually alternates between the
values 1.6 and 2.7. This type of behavior cannot be found in solutions of linear
equations. The solutions of nonlinear equations show a much richer variety of
behaviors. Another important difference is that linear equations admit closed form
solutions whereas nonlinear equations typically cannot be solved analytically.

EXAMPLE 6.2 Population Growth

Discrete dynamical systems are widely used in population modeling, in particular
for species which have no overlap between successive generations and for which
births occur in regular, well-defined ‘breeding seasons’. Let pn be the average
population of a species between times nτ and (n + 1)τ . The time step τ depends
on the particular species and can range from an hour to several years. For example
many species of bamboo grow vegetatively for 20 years before flowering and then
dying.
In population dynamics one constructs a model for the change ∆pn = pn+1 −
pn . The simplest model is a linear model, ∆p = kpn + β, where k is called the
reproduction rate and β models constant immigration (β > 0) or emigration (β <
0). The difference equation that results from this model assumption,

pn+1 − pn = kpn + β,
 Section 6.1 Introduction 109

2 3

2.5
1.5
2

xn
n
1 1.5

x 2
x
n+1
=2.8*x −x
n n
1 x =3.3*x −x2
n+1 n n
0.5
x =0.1 0.5
0 x0=0.1

0 0
0 5 10 n 15 20 25 0 5 10 n 15 20 25

(a) (b)

FIGURE 6.2: Numerical solutions for equation (6.4).

is again of the form of Equation (6.1).

Competition for resources usually leads to nonlinear difference equations. We
will see that the simplest model that takes competition into account leads to the
equation
pn+1 = rpn − p2n , (6.4)
which is of the form of Equation (6.3). Equation (6.4) is known in the literature as
logistic map. Its prominent feature are very complicated, so called chaotic solutions
in certain ranges of the parameter r.

The equation
xn+2 + 2xn+1 + 3xn = cos(n)
is an example of a linear second order difference equation. We shall see that this
type of equation always can be transformed to a linear system of two first order
equations. The general form of such a system is

xn+1 = axn + byn + fn ,

yn+1 = cxn + dyn + gn ,

where fn , gn are known sequences. If the right hand sides are replaced by nonlinear
functions we have a nonlinear system, for instance

xn+1 = axn − bx2n − cxn yn ,

yn+1 = dxn − eyn2 − f xn yn .

This system is used in population modeling as a model for the population growth
of two interacting species. The terms −bx2n and −eyn2 model competition within
each of the two species whereas the terms −cxn yn and −f xn yn model competition
between the species.
 110 Chapter 6 Modeling with Discrete Dynamical Systems

6.2 LINEAR FIRST ORDER DIFFERENCE EQUATIONS

6.2.1 Analytical Solutions
Possibly the simplest nontrivial difference equation has the form

xn+1 = axn . (6.5)

This equation has the special solution xn = 0. Since it is constant it is said to be

an equilibrium solution. The value of the constant, x = 0, is called an equilibrium
value or shortly an equilibrium. The solutions for initial values x0 6= 0 are found
by implementing the iteration,

x1 = ax0
x2 = ax1 = a2 x0
x3 = ax2 = a3 x0
..
.
xn = an x0 . (6.6)

From (6.6) we can easily infer how the qualitative behavior of xn depends on a: if
|a| > 1 then xn goes off to infinity (the equilibrium is said to be unstable), whereas
if |a| < 1 then xn tends to 0 (the equilibrium is said to be stable). This explains
the behavior of the numerical solutions of Figures 6.1 (a) and (c). Note also that
if a > 0 then xn has the same sign as x0 for all n. In contrast if a < 0 the solution
alternates between positive and negative values.
The cases a = 1 and a = −1 are special. If a = 1 we have xn = x0 for all n,
hence every x is an equilibrium. If a = −1 the solution xn = (−1)n x0 flips back
and forth between x0 and −x0 .
A more general equation is the following,

xn+1 = axn + b. (6.7)

An equilibrium is determined by xn+1 = xn = x for all n, hence

b
x = ax + b ⇒ x = ,
1−a
where we assume a 6= 1. We can transform (6.7) to (6.5) by subtracting the
equilibrium. Set
yn = xn − x.
Then

yn+1 = xn+1 − x = axn + b − x

= a(yn + x) + b − x = ayn ,

and so yn = an y0 . The solution of (6.7) is found by transforming yn back to xn ,

b b
xn = yn + x = axn (x0 − x) + x = an (x0 − )+ . (6.8)
1−a 1−a
 Section 6.2 Linear First Order Difference Equations 111

Again the value of |a| determines whether xn goes off to infinity or approaches x,
and the sign of a determines whether xn − x alternates or has a constant sign.

EXAMPLE 6.3
The equation
1
xn+1 = xn + 1
2
is of the form of (6.7). The equilibrium is
b
x= = 2.
1−a
Since a = 1/2 < 1 and a > 0 the solutions approach the equilibrium 2 and the sign
of xn − 2 is the same for all n. This explains the behavior of the numerical solutions
shown in Figure 6.1 (d).

A more general form than (6.7) is provided by the equation

xn+1 = axn + bn , (6.9)
where bn is a given sequence. This equation is said to be nonhomogeneous due to
the presence of the bn term. If bn = 0 for all n, (6.9) simplifies to (6.5) and then
the equation is called homogeneous. We refer to (6.5) as the homogeneous equation
associated with (6.9). In the special case in which bn = b = const we were able to
transform the nonhomogeneous equation to its associated homogeneous equation,
but if bn varies with n this is no longer possible.
Definition 4. A one parameter family of solutions of (6.9) is an expression
xn = xn (c) that depends linearly on a parameter c and satisfies (6.9) iden-
tically in n and c. A particular solution is a solution that contains no free
parameters. A one parameter family of solutions is a general solution if for
every particular solution pn we can find a value c of c such that pn = xn (c)
for all n.
Consider now the difference hn = qn − pn of two particular solutions qn and
pn of (6.9). The computation
hn+1 = qn+1 − pn+1 = (aqn + bn ) − (apn + bn )
= a(qn − pn ) = ahn
shows that hn is a solution of the homogeneous equation (6.5). Since h0 = q0 − p0
it follows from (6.6) that hn = (q0 − p0 )an and so,
qn = (q0 − p0 )an + pn .
If we assume pn is a known particular solution, this equation allows to find any
other particular solution qn from its initial value q0 . Thus if we write
xn = can + pn , (6.10)
and consider c as parameter, the solution qn is simply obtained by setting c = q0 −p0 .
We therefore have proved the following theorem.
112 Chapter 6 Modeling with Discrete Dynamical Systems

Theorem 5. Let pn be a particular solution of the nonhomogeneous equation

xn+1 = axn + bn .
Then the family
xn = can + bn
is a general solution.
Note that there is no unique general solution. For instance,
xn = can + (pn + 5an )
is also a general solution because pn + 5an is another particular solution.

EXAMPLE 6.4
Verify that pn = −n − 1 is a particular solution of
xn+1 = 3xn + 2n + 1.

Solution To test that an expression is a solution of a difference equation we just

have to plug it into the equation and check if both sides are the same. Now the left
hand side evaluates to
pn+1 = −(n + 1) − 1 = −n − 2,
and the right hand side to
3pn + 2n + 1 = 3(−n − 1) + 2n + 1 = −n − 2.
Since these are the same we have verified that pn is a solution. It is a particular
solution because it does not depend on parameters.

EXAMPLE 6.5
Find the general solution of
xn+1 = 3xn + 2n + 1
and the particular solution that satisfies x0 = 1.
Solution Form Example 6.4 we know that pn = −n − 1 is a particular solution.
Since a = 3 the general solution is
xn = c3n − n − 1.
To find the particular solution asked for we evaluate at n = 0,
x0 = c − 1 = 1.
It follows that c = 2, hence
xn = 2 3n − n − 1
is the solution with x0 = 1.
 Section 6.2 Linear First Order Difference Equations 113

bn form of particular solution conditions

(6.11) pn = (A0 + A1 n + · · · + AM nM )bn b 6= a
pn = n(A0 + A1 n + · · · + AM nM )bn b=a
(6.12) pn = (A0 + A1 n + · · · + AM nM )bn cos(kn) k 6= 0, π
+ (B0 + B1 n + · · · + BM nM )bn sin(kn)

TABLE 6.1: Solution forms pn for bn given by Equations (6.11) and (6.12)

To complete the solution of the nonhomogeneous equation (6.9) we need to

find a particular solution. For general terms bn this can be a complicated task.
However there is a method that applies always if bn is a combination of powers of n
(n0 , n1 , n2 etc.), trigonometric functions of n, and powers bn . This method is called
method of undetermined coefficients.

Method of undetermined coefficients. Assume bn has one of the following

forms,
bn = (c0 + c1 n + . . . + cM nM )bn , (6.11)
where cM 6= 0, or

bn = (c0 + c1 n + · · · + cM nM )bn cos(kn)

+ (d0 + d1 n + . . . + dM nM )bn sin(kn), (6.12)

where at least one of cM or dM is nonzero. The coefficients b, k and cj , dj (0 ≤ j ≤

M ) are assumed to be given numbers. It can be shown that if bn is as in (6.11) or
(6.12), then there exists a unique particular solution pn of the form as summarized
in Table 6.2.1. To find the values of the coefficients Aj , Bj (0 ≤ j ≤ M ), one sets up
a trial form for pn according to the table with initially undetermined values of the
coefficients, substitutes the trial form into the difference equation, and determines
the values of the coefficients from the condition that pn be a solution. If bn is a
linear combination of several terms of the form of (6.11) or (6.12), with different
values of b or (b, k), each of them can be treated separately and the results are
added up.

EXAMPLE 6.6
Find a particular solution of

xn+1 = 3xn + 2n + 1.

Solution Here bn = 2n + 1 is of the form (6.11) with b = 1 and M = 1. Thus we

use pn = A + Bn as trial form and substitute this into the difference equation to
obtain,
A + B(n + 1) = 3(A + Bn) + 2n + 1,
or
(2A − B + 1) + (2B + 2)n = 0.
114 Chapter 6 Modeling with Discrete Dynamical Systems

This equation holds for all n if A and B satisfy the equations 2A − B = −1 and
2B = −2. The solution is A = B = −1, hence

pn = −n − 1.

EXAMPLE 6.7
Find a particular solution of

xn+1 = −xn + cos 2n.

Solution Substitution of the trial form pn = A cos 2n+B sin 2n into the difference
equation yields

A cos 2(n + 1) + B sin 2(n + 1) = −A cos 2n − B sin 2n + cos 2n.

We apply the formulae for cos(α + β) and sin(α + β) to the terms on the left hand
side and then rearrange the equation as

[A(1 + cos 2) + B sin 2 − 1] cos 2n + [−A sin 2 + B(1 + cos 2)] sin 2n = 0.

This equation holds for all n if the terms in both brackets vanish. Setting these
terms equal to zero gives the following system of equations for A and B,

A(1 + cos 2) + B sin 2 − 1 = 0

−A sin 2 + B(1 + cos 2) = 0,

with the solution

1 sin 2
A= , B= .
2 2(1 + cos 2)
Hence the particular solution is

1 sin 2 cos 2n cos 2n + cos 2(n − 1)

pn = cos 2n + = .
2 2(1 + cos 2) 2(1 + cos 2)

EXAMPLE 6.8
Find a particular solution of

xn+1 = xn /2 + n(1/2)n .
 Section 6.2 Linear First Order Difference Equations 115

Solution Here a = b = 1/2, so the trial function is pn = n(An + B)(1/2)n . Again

we substitute pn into the difference equation,

[A(n + 1)2 + B(n + 1)](1/2)n+1 = (An2 + Bn)(1/2)n /2 + n(1/2)n .

We multiply this equation by 2n+1 and rearrange terms as

2(A − 1)n + (A + B) = 0.

Thus A = −B = 1 and the particular solution is

pn = (n2 − n)(1/2)n .

6.2.2 Modeling Examples

(A) Savings Accounts and Loans

Savings Accounts. Assume you open a savings account at an annual interest

rate of r% and with monthly compound of interest. Let an be the dollar amount
on the account at the end of month n after the opening date. The amount at the
end of month n + 1 is
an+1 = an + in + pn ,
where pn is the total deposit (withdrawal if pn < 0) and in is the interest earned,
 
r 1
in = an .
100 12

Thus an satisfies the nonhomogeneous, linear first order difference equation,

an+1 = kan + pn , (6.13)

where
r
k =1+ .
1200
If pn = p = const we know the solution already (Equation (6.8) with a = k, b = p,
xn = an ),
p p (k n − 1)p
an = k n (a0 + )− = k n a0 + . (6.14)
k−1 k−1 k−1

EXAMPLE 6.9
After graduating from High School Peter works for four years. During this time he
deposits each month $1000 on a savings account at an annual interest rate of 5%
(no initial deposit). The next four years Peter spends on College. During this time
he withdraws each month an amount of $pw from his savings account so that at
the end of the four years the balance is zero again. Find pw and the total interest
earned.
116 Chapter 6 Modeling with Discrete Dynamical Systems

Solution Letting p be the the monthly deposit, the accumulated amount on

Peter’s savings account after the first four years is

(k 48 − 1)p
a48 = .
k−1
After the second four years this has evolved into

(k 48 − 1)pw k 48 − 1 48
a96 = k 48 a48 − = (k p − pw ).
k−1 k−1

Solving the equation a96 = 0 for pw gives pw = k 48 p. With p = $1000 and k =

1+5/1200 this evaluates to pw = $1220.89. The total interest earned is 48(pw −p) =
$10, 602.72.

Loans. Equation (6.13) also holds for loans. In this case a0 is the amount bor-
rowed and an is the amount owed after n months. The term −pn > 0 is the monthly
payment. For constant monthly payment p the difference equation for an is

an+1 = kan − p, (6.15)

with the solution

(k n − 1)p
an = kan0 − .
k−1
Note that (6.15) has an unstable equilibrium a = p/(k − 1). If a0 > a the solution
grows without bound when n increases. While for savings accounts this may be
desirable, it is certainly not tolerable for loans.
The term of a loan is the time N (in months) when the loan is paid off. Setting
aN = 0 leads to a linear relation between monthly payment and initial debt,

k N (k − 1)
p= a0 . (6.16)
kN − 1

EXAMPLE 6.10
You decide to purchase a home with a mortgage at 6% annual interest and with a
term of 30 years. For k = 1 + 6/1200 = 1.005 and N = 360 the factor

k N (k − 1)
R=
kN − 1

in Equation (6.16) is R = 0.00600. If the house costs a0 = $200, 000, the monthly
payment is p = Ra0 = $1, 199.10. On the other hand, if your income restricts your
monthly payment to a maximum of pm = $1000, the maximal amount you can
spend for the house is pm /R = $166, 791.61.
 Section 6.2 Linear First Order Difference Equations 117

If p, a0 and k are fixed, the equation (6.16) may be considered as an equation

for the term N . Writing k N = eN ln k , Equation (6.16) can be rewritten as
p
eN ln k = ,
p − (k − 1)a0

hence
− ln[1 − (k − 1)a0 /p]
N= . (6.17)
ln k
Note however that the right hand side of (6.17) needs not to be an integer. Nev-
ertheless it can be used to estimate N and then to improve p or a0 . For example,
assume you need $200, 000 and you want your payment to be close to, but not above
$1500. With r = 8%, a0 = 200, 000 and p = 1500, (6.17) evaluates to N = 330.68.
If this is rounded up to N = 331, Equation (6.16) gives p = $1499.60.
In our last example on savings accounts and loans we have to solve the non-
homogeneous equation (6.9) with nonconstant bn .

EXAMPLE 6.11
An employee starts her position at the age of 25 with an annual salary of $40, 000.
She deposits each month 8% of her monthly salary on a retirement savings account.
The salary increases by 3% each year and the annual interest rate of her retirement
savings account is 6%. What is the accumulated amount when she retires at the
age of 65?
Solution Let Am be the accumulated amount on the retirement savings account
at the end of year m and let am,n be the accumulated amount in month n of year
m + 1, that is,
am,0 = Am , am,12 = Am+1 .
The amount am,n satisfies difference equation

am,n+1 = kr am,n + kp sm , (6.18)

where kr = 1 + 6/1200 = 1.005, kp = 8/1200 and sm is the salary in year m + 1.

The salary satisfies the homogeneous difference equation

sm+1 = ks sm ,

with ks = 1 + 3/100 = 1.03 and s0 = 40, 000, hence

sm = ksm s0 .

The solution of (6.18) is

(krn − 1)ksm kp s0
am,n = krn am,0 + .
kr − 1
Evaluating this at n = 12 yields

Am+1 = ka Am + f ksm , (6.19)

118 Chapter 6 Modeling with Discrete Dynamical Systems

where
(kr12 − 1)kp s0
f= = $3289.48, ka = kr12 = 1.0616778.
kr − 1
It remains to solve the nonhomogenous difference equation (6.19). By using the
method of undetermined coefficients a particular solution can be determined to be
pm = f ksm /(ks − ka ). The solution with initial value A0 = 0 then is

(ksm − kam )f
Am = .
ks − ka
For m = 40 this evaluates to A40 = 799, 106.39. Hence the employee starts her
retirement with an amount of $799, 106.39 on her retirement savings account.

(B) Cooling and Heating

Newton’s law of cooling states that the rate of change of the temperature of an
object is proportional to the difference of the temperature of the object and its
surrounding. Let ∆Tn = Tn+1 − Tn be the change in temperature of the object
over a time interval τ , typically τ = 1 hour. According to Newton’s law of cooling
we have
∆Tn ∝ Rn − Tn ,
or
∆Tn = k(Rn − Tn ),
where Rn is the surrounding temperature. Since we know that temperature de-
creases if Rn > Tn it follows that k > 0. The difference equation that arises from
this model is
Tn+1 = Tn + k(Rn − Tn ).
If Rn = R = const this equation is again of the form (6.7) with solution

Tn = (1 − k)n (T0 − R) + R.

Note that the equilibrium solution is Tn = R as expected. The equilibrium is stable

if 0 < k < 2. However if 1 < k < 2 the temperature would oscillate about the
surrounding temperature which does not make sense physically, hence 0 < k < 1.

EXAMPLE 6.12
A murder victim is discovered in an office building that is maintained at 68 degrees
F. Given the medical examiner found the body temperature to be 88 degrees F at
8am and that one hour later the body temperature was 86 degrees F, at what time
was the crime committed?
Solution Setting T0 = 98.6 (where 0 is the time the crime was committed) and
R = 68 we obtain
Tn = 68 + 30.6(1 − k)n .
 Section 6.3 Linear Second Order Equations 119

If we define n1 as the time the body was observed initially by the medical examiner
and the time one hour later as n1 + 1 we have the equations

Tn1 = 88 = 68 + 30.6(1 − k)n1 ,

and
Tn1 +1 = 86 = 68 + 30.6(1 − k)n1 +1 .
These two equations may be solved to give k = 1/10 and n1 = 4.036. So the crime
was committed just before 4am.

6.3 LINEAR SECOND ORDER EQUATIONS

6.3.1 Homogeneous Equations
We begin by considering the second order linear homogeneous difference equation

xn+2 + αxn+1 + βxn = 0 (6.20)

It is readily verified that this equation has solutions of the form

xn = λn

Upon substitution into Equation (6.20) we obtain the auxiliary equation

λ2 + αλ + β = 0

This quadratic equation has solutions that break down into three cases: i) both
solutions real and distinct, ii) one real double solution, and iii) a pair of complex
solutions as p
−α ± α2 − 4β
λ± =
2

Case i: α2 − 4β > 0. Two real roots.

In this case p
−α + α2 − 4β
λ+ =
2
and p
−α − α2 − 4β
λ− =
2
are both real and the solution is

hn = c1 (λ+ )n + c2 (λ− )n

Since the equation is linear we know that the superposition of solutions is again a
solution. Notice that there are now two free parameters c1 and c2 to accommodate
the two initial conditions x0 and x1 required for a second order difference equation.
Notice also that hn → 0 for n → ∞ if λ± | < 1, but in general |hn | → ∞ if |λ+ | > 1.
120 Chapter 6 Modeling with Discrete Dynamical Systems

EXAMPLE 6.13
xn+2 = xn+1 + xn
The auxiliary equation is now
λ2 − λ − 1 = 0
The solutions to this quadratic are
√
1± 5
λ± =
2
Thus, the general solution to the homogeneous problem is
 √ n  √ n
1+ 5 1− 5
hn = c1 + c2
2 2
If we select h0 = h1 = 1 we have the Fibonocci sequence {1, 1, 2, 3, 5, 8, 13, . . . }. Em-
ploying this pair of initial conditions it is easily shown that the particular solution
is √  √ n  √  √ n
5+1 1+ 5 5−1 1− 5
hn = √ + √
2 5 2 2 5 2
You might impress your friends by telling them the 50th number in this sequence
h50 = 20365011074. It is also apparent that these numbers increase exponentially
fast.

Case ii: α2 − 4β = 0. One real (double) root.

In this case
α
λ+ = λ− = −
2
so we only have one solution while we require two for the general solution of a
second order difference equation.
It is not hard to verify that in this instance the second solution is actually
−α n
xn = n( )
2
. (See Exercise 6.15). Now the general solution to this homogeneous equation is
α
n α
n
hn = c1 − + c2 n −
2 2

EXAMPLE 6.14
xn+2 + 2xn+1 + xn = 0
The auxiliary equation is
λ2 + 2λ + 1
which has the solution λ = −1. Thus, the general solution to this homogeneous
problem is
hn = c1 (−1)n + c2 n(−1)n
 Section 6.3 Linear Second Order Equations 121

Case iii: α2 − 4β < 0. Two complex roots.

The solution to the auxiliary equation is again
p
−α ± α2 − 4β
λ± =
2
Based on the fact α2 − 4β < 0 we rewrite this as
p
−α 4β − α2
λ± = ±i
2 2
√
where i = −1.1
We could now write the solution
 p n  p n
−α 4β − α2 −α 4β − α2
hn = c1 +i + c1 −i
2 2 2 2
but this form would not provide much insight. Instead we employ Demoivre’s
theorem that states
exp(inx) = cos(nx) + i sin(nx)
To exploit this formula we need to recall that each solution to the auxiliary equation
can be written in its complex polar form

z = x + iy = r exp(iθ)

where x = r cos θ and y = r sin θ. Thus, we take

p
−α 4β − α2
x= , and y =
2 2
To compute the polar form we need r and θ. Recall

r2 = x2 + y 2

so
p
2 −α 2 4β − α2 2
r =( ) +( )
2 2
=β

So p
r= β
The angle satisfies p
y 4β − α2
tan θ = =
x −α
In polar form, the solution is

hn = c1 rn exp(inθ) + c2 rn exp(−inθ)
1 Unlike the previous cases we now assume familiarity with basic complex numbers.
 122 Chapter 6 Modeling with Discrete Dynamical Systems

The associated real form is

hn = rn (c1 cos(nθ) + c2 sin(nθ),

where we have used the facts that exp(inθ) = cos(nθ) + i sin(nθ) and that the real
and imaginary parts of a complex solution are real solutions (see problems). The
form of the solution tells that hn → 0 for n → ∞ if r < 1 and |hn | → ∞ if r > 1.
If r = 1 the solution remains bounded, but does not approach zero.

EXAMPLE 6.15
Find the general solution to the homogeneous difference equation

xn+2 + 2xn+1 + 5xn = 0

The auxiliary equation gives the solutions

λ± = −2 ± i

If we write these in polar form we have

hn = 5n/2 (c1 exp(inθ) + c2 exp(−inθ))

where tan θ = 1/2. The associated real valued form is

hn = 5n/2 (c1 cos(nθ) + c2 sin(nθ)).

6.3.2 The Cobweb Model Revisited

Consider a supply curve
p = ms q + b s
and a demand curve
p = md q + b d
Here we derive a formula for the values (qn , pn ) that are the iterations along the
supply and demand curves that either converge to an economic equilibrium or spiral
out of control. Let the starting point on the demand curve be (q0 , p0 ). The next
iteration is then given by
p0 − bs
(q1 , p1 ) = ( , p0 )
ms
Similarly,
(q2 , p2 ) = (q1 , md q1 + bd ),
p2 − bs
(q3 , p3 ) = ( , p2 )
ms
and
(q4 , p4 ) = (q3 , md q3 + bd )
Thus, we have established the following pattern:
 Section 6.3 Linear Second Order Equations 123

(q2n , p2n ) = (q2n−1 , md q2n−1 + bd )

and
p2n − bs
(q2n+1 , p2n+1 ) = ( , p2n )
ms
It is now possible to create a second order difference equation for both qn and
pn . Since
p2n − bs
q2n+1 =
ms
it follows, upon substituting for p2n that

(md q2n−1 + bd ) − bs
q2n+1 =
ms
or,
md bd − b s
q2n+1 = q2n−1 + . (6.21)
ms ms

A Nonhomogeneous Second Order Equation. The equation (6.21) is of the

form
q2n+1 = αq2n−1 + β
This is a nonhomogeneous second order difference equation whose general solution
is, as in the first order case, given by

xn = hn + pn ,

where hn is the general solution of the associated homogeneous equation and pn is

a particular solution of the nonhomogeneous equation.
The associated homogeneous equation is

q2n+1 = αq2n−1

and has the auxiliary equation

λ2 = α
so the general solution to the homogeneous problem is

hn = c1 αn/2 + c2 (−α1/2 )n

As the nonhomogeneous term is a constant we first search for a particular

solution of the form pn = A. This must be an equilibrium solution, if it exists.
Solving for A then gives
A = αA + β
or
β
A=
1−α
 124 Chapter 6 Modeling with Discrete Dynamical Systems

In terms of the original variables of the supply and demand problem, α =

md /ms , β = (bd − bs )/ms , the general solution to the nonhomogeneous equation
now becomes
 n/2  n
md md 1/2 bd − b s
qn = c1 + c2 −( ) +
ms ms ms − m d

It is clear from our previous work that this equation will only converge if
md
| |<1
ms
Note also that if this condition holds then the quantity supplied converges,

bd − b s
qn →
ms − m d
and approaches the market equilibrium.

6.4 NONLINEAR DIFFERENCE EQUATIONS AND SYSTEMS IN POPULATION

MODELING
In this section we will consider a sequence of modifications of a population model
that characterize the modeling process and illustrate how including or deleting
terms in equations can have dramatic effects on the predictive powers of a model.
The simplest model for population growth makes the assumption that there is
no competition for resources such as nutrients or habitat. This exponential growth
is readily captured by the simple difference equation

pn+1 − pn = ∆pn = kpn (6.22)

where the growth constant k > 0 reflects the rate of reproduction. One would
assume that for rabbits this constant would be larger than for elephants. Actual
values for k must be determined empirically from the data using a data fitting
technique such as least squares.
If instead of simply taking k > 0 in Equation (6.22) we could have modeled
both the birth rate kb and the death rate kd such that

pn+1 − pn = kb pn − kd pn (6.23)

Clearly now we may write

k = kb − kd
and as we would expect, if k > 0 the model predicts that the population grows
exponentially fast and if −1 < k < 0 then the population decays exponentially fast.
Values of k in the rang k < −1 do not make sense because then the solution would
oscillate between positive and negative values.
The effect of adding to a population via immigration or subtracting via emi-
gration is captured by
pn+1 − pn = kpn + βn (6.24)
 Section 6.4 Nonlinear Difference Equations and Systems in Population Modeling 125

where βn is the net flux of population. Now we might expect that growth rates
could be offset by immigration or emigration. For example k < 0 but βn = β can
produce a positive equilibrium population.
Obviously ignoring competition for finite resources places significant limita-
tions on this model. It will work well where the assumptions hold true but when
the effects of competition for resources become important it will not capture them.
To model competition we may argue as follows: competition occurs when there is
interaction between two members of a species and the total amount of competition
is the number of ways we can select subsets of 2 from a population p which is

p(p − 1)
number of pairwise interactions ∝
2
where we have divided by two to compute the number of combinations rather than
permutations. Now we may modify the model to incorporate competition as

pn+1 − pn = k1 pn − k2 pn (pn − 1) (6.25)

again ignoring effects due to migration. Here we are assuming k2 > 0 and use
the negative sign to reflect the fact that competition reduces the population. This
equation can be simplified to

pn+1 − pn = c1 pn − c2 p2n (6.26)

This is the well-known logistic difference equation for population growth and it
appears to correspond well to the growth of bacteria in agar jelly, for example.
Superficially we see that the difference between the model that does not model
competition and the one that does is a quadratic term. A more fundamental dif-
ference is that Equation (6.22) is linear while Equation (6.26) is nonlinear. The
only fixed point for Equation (6.22) is p = 0. For Equation (6.26) there are now
two fixed points p1 = 0 and p2 = cc12 ; see Figure 6.3. From the plot of pn+1 − pn
it is clear that this new model predicts that the population will be limited, i.e., it
can’t grow unbounded to ∞ because as soon as pn > c1 /c2 then ∆pn < 0 so the
population must decrease.
One is initially tempted to conclude that the equilibrium point p1 = 0 is
unstable while the equilibrium point p2 = c1 /c2 is stable. As we shall see in
the numerical simulations this can be true, but for certain values of c1 and c2
the situation can be much more complicated including periodic and even chaotic
solutions!

6.4.1 Systems of Equations and Competing Species

Now consider two species A and B whose populations are denoted an and bn ,
respectively. If we assume that these species have infinite resources and compete
neither with themselves or each other then we would propose the simple system of
difference equations
an+1 − an = k1 an
bn+1 − bn = k2 bn
126 Chapter 6 Modeling with Discrete Dynamical Systems

pn+1−pn>0

pn+1−pn<0
pn+1−pn

0
pn c /c
1 2

pn+1−pn>0

pn+1−pn <0

FIGURE 6.3: A plot of the change in population pn+1 − pn as a function of the population pn .

This system is said to be uncoupled as the values of an do not influence bn and,

similarly, the values of bn do not influence an .
If species B eats the same kind of food species A does, but species A does not
eat the same kind of food species B does we have the model

an+1 − an = g1 an − c1 an bn
bn+1 − bn = g2 bn

If species A and B both like each others food we would employ the model

an+1 − an = g1 an − c1 an bn
bn+1 − bn = g2 bn − c2 an bn

See Figure 6.4 for a numerical simulation of this system. Note that this nonlinear
system does not have a closed form solution. For the parameters selected we see
that even though species A initially has a lower population it appears to grow
without bound while population B becomes extinct. Here we may conclude that
species A is more fit than species B and consequently survives.
If species A and B compete both with each other and with themselves the
population model would then become

an+1 − an = g1 an − c1 an bn − k1 a2n
bn+1 − bn = g2 bn − c2 an bn − k2 b2n
 Section 6.4 Nonlinear Difference Equations and Systems in Population Modeling 127

35

30 a
n
b
n
25

20

15

10

0
0 10 20 30 40 50
iteration n
FIGURE 6.4: Competition for finite resources. We have selected the initial conditions a1 = 25 and
b1 = 30. In addition the parameters were chosen to be g1 = .047, c1 = .012, g2 = .023, c2 = .015.

Notice that if population B becomes extinct while A survives then the model reduces
to the logistic difference equation for a single species.

Predator Prey Model. Now consider modeling the interaction between

natural predators and their prey. A classic example of this relationship is given by
foxes and rabbits. The population of foxes and rabbits are intimately linked given
that the rabbits are the food supply for the foxes. When the population of rabbits
increases one can predict an associated, though possibly time lagged, increase in
the number of foxes. Conversely, when the number of rabbits decreases the less
food there is for the foxes. Of course an increase in the number of foxes will result
in more rabbits being eaten and thus a reduction in the rabbit population.
Let’s develop a model for this situation. First, denote the fox population by
fn and the rabbit population by rn . If we assume that in the absence of rabbits
the fox population becomes extinct we have the model

∆fn = −g1 fn

where the constant g1 > 0. If rabbits are available, then they should contribute
positively to a change in the fox population. It seems reasonable to assume that
the increase in the fox population will be proportional to the number of fox and
 128 Chapter 6 Modeling with Discrete Dynamical Systems

rabbit interactions which is given by the product fn rn . Thus, in the presence of

rabbits we may model the change in the fox population to be

∆fn = −g1 fn + c1 fn rn

where the constant c1 > 0

Now the rabbits should multiply in the absence of foxes

∆rn = g2 rn

where the constant g2 > 0. The impact of the foxes on the rabbits is presumably
also proportional to the number of interactions but now this reduces the rabbit
population.
∆rn = g2 rn − c2 fn rn
In summary we have the model

fn+1 = (1 − g1 )fn + c1 fn rn (6.27)

rn+1 = (1 + g2 )rn − c2 fn rn (6.28)

Note that we have omitted the competition amongst the foxes for the rabbits as
well as the competition amongst the rabbits for their food. This is easily captured
by extending the above system to

fn+1 = (1 − g1 )fn + c1 fn rn − d1 fn2 (6.29)

rn+1 = (1 + g2 )rn − c2 fn rn − d2 rn2 (6.30)

See Figure 6.5 for a simulation of the above equations. Note that the predicted
oscillation is in fact there, however it is damped and the solution proceeds to a
stable equilibrium.

6.5 EMPIRICAL MODELING

One may imagine that true observations, e.g., from populations in nature, will not
be precise due to limitations in counting species in the wild. Thus, the data will
contain what we refer to as a unknown noise component. In general, model selection
can be arrived at by

1. Collect observations to build models

2. Propose models, e.g., predator prey or competing species
3. Compute model coefficients in each case
4. Compare models through validation and testing

Now we present the method of least squares as a means to determine our

unknown model coefficients.

6.5.1 Non-Newtonian Fish?

Recall that Newton’s Law of Cooling states that the temperature change in a body
is proportional to the difference between the temperature of the body Tn and the
 Section 6.5 Empirical Modeling 129

2000

1800
rabbits
foxes
1600

1400

1200
population

1000

800

600

400

200

0
0 100 200 300 400 500 600 700 800
iteration

FIGURE 6.5: Simulation of preditor prey equations. We have selected the initial conditions f1 = 25 and
r1 = 100. In addition the parameters were chosen to be g1 = 0.01, c1 = 0.0001, g2 = 0.1, c2 = 0.0005,
d1 = 0.0001 and d2 = 0.

surrounding temperature M , i.e., as a difference equation

∆Tn = k(M − Tn )

After repeatedly overcooking a certain kind of fish based on this law a frustrated
cook has decided to take science into her own hands. She speculates that the actual
law of cooking for this fish has the more general form

∆Tn = k(M − Tn )α

and that for certain types of foods, call them Non-Newtonian foods, that α 6= 1.
To test her hypothesis, our cook measures the temperature of a fish every
minute until it approaches the temperature of the oven which is set to 425 degrees
F. The results of her data collection are shown in Figure 6.6.
Now ∆Tn is known since Tn is known for n = 1, . . . , 200. Thus, for any α and
k we can compute a model error of
X
2
E(α, k) = ∆Tn − k(425 − Tn )α
n

We recall from our previous work with least squares that computing α and k requires
differentiating the error term E(α, k) with respect to α and k. For this particular
130 Chapter 6 Modeling with Discrete Dynamical Systems

Non−Newtonian Fish
450

400

350

300
Tn: temperature of fish

250

200

150

100

50

0
0 20 40 60 80 100 120 140 160 180 200
time n in minutes

FIGURE 6.6: Observations of a Non-Newtonian fish. These are (synthetic) measurements of the
temperature of the fish as a function of time.

model it is simpler to employ a logarithmic transformation

yn = ln ∆Tn
b = ln k
xn = ln(425 − Tn )
giving X
E(α, b) = (yn − b − αxn )2
n
Differentiating these with respect to α and b and setting the results equal to zero
produces the equations
 P 2 P    P 
n xn n x n b n y n xn
P = P
n xn P α n yn

Solving these equations using only the first 101 observations T0 , T1 , . . . , T100 and
the MATLAB code provided produces the results
α = 1.25
and
k = 0.01
 Section 6.5 Empirical Modeling 131

160

140

120

100
population

80

60

rabbits
40 foxes

20

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
iteration x 10
4

FIGURE 6.7: Simulation of preditor prey equations. We have selected the initial conditions f0 = 25
and r0 = 100. In addition the parameters were chosen to be g1 = .01, g2 = .0005, c1 = .0001, c2 =
.0001, d1 = 0.0, d2 = 0.0

6.5.2 Predator or Prey?

Assume that the data in Figure 6.7 is provided. The goal is to see if we can calculate
the coefficients of the preditor prey equations that will reproduce this data. Thus,
given the tentative model

∆fn = −g1 fn + c1 fn rn
∆rn = g2 rn − c2 fn rn

the points {fn , rn } are now observations while the equation coefficients {g1 , g2 , c1 , c2 }
are to be determined.
The least squares error is now
X X
E(g1 , c1 , g2 , c2 ) = (∆fn + g1 fn − c1 fn rn )2 + (∆rn − g2 fn + c2 fn rn )2 (6.31)
n n

Setting
∂E ∂E ∂E ∂E
= = = =0
∂g1 ∂c1 ∂g2 ∂c2
produces the necessary conditions for a minimum error. Taking the uncoupled
132 Chapter 6 Modeling with Discrete Dynamical Systems

equations for c1 and g1 we have


−P n fn2
P P 2    P 
n fn rn g1
= n (∆fn )fn
(6.32)
− n fn2 rn 2 2
P P
n fn rn c1 n (∆fn )fn rn

These must be solved simultaneously with the uncoupled conditions for c2

and g2 , i.e.,
 P 2
− n rn − Pn rn2 fn
P    P 
g2 n (∆rn )rn
2 = (6.33)
− n fn2 rn2
P P
n rn fn c2 n (∆rn )fn rn

Solving these equations produces the exact coefficients that were used to gen-
erate the data. In principal, this procedure may be applied to direct observations
from nature. One may conclude if a model fits the data and, if so, which species
plays which role, i.e., by examining the computed signs of c1 , c2 , g1 and g2 one may
infer which species is the predator and which is the prey. See the MATLAB code
for these equations in the Appendix.
 Section 6.5 Empirical Modeling 133

PROBLEMS
6.1. Consider the following equations and identify as
• linear or nonlinear
• homogeneous or nonhomogenous
• which order
(a) x2n+1 + xn = 1.
(b) xn+1 = xn−1 + 2
(c) xn+1 = sin(xn−1 )
(d) xn+3 = xn+1 + xn−3 + n2
6.2. Determine particular solutions to the following equations
(a) xn+1 = xn + 1
(b) xn+1 = 5xn + n2
(c) xn+1 = x2n + 6n
6.3. Show that the real and imaginary parts of a complex solution to a linear difference
equation are also solutions to the same difference equation.
6.4. Determine general solutions to the following equations
(a) xn+1 = xn + 1
(b) xn+1 = 5xn + n2
(c) xn+1 = x2n + 6n
(d) xn+1 = x2n + 4n2 + 2n + 1
6.5. You currently have $5000 in a savings account that pays 6% interest per year.
Interest is compounded monthly. You add another $200 each month. What do
you have on your savings account after five years, and what is the total interest
earned during these five years?
6.6. You owe $500 on a credit card that charges 1.5% interest in each month. You
can pay $50 each month and make no new charges. When is your loan paid off
and what is your last payment? How much interest have you paid?
6.7. Your parents are considering a 30–year $100, 000 mortgage at an annual interest
rate of 6%. What is the monthly payment, and what will be the total interest
paid?
6.8. Mary receives $5000 as graduation present from her parents when graduating
from High School. She deposits the money on a savings account at an annual
interest rate of 3%. Interest is compounded monthly. Before going to college she
works for three years, and during this time she deposits each month a certain
amount on the savings account. She plans to withdraw $1200 each month in
her first year on college, and to increase the monthly withdrawal in each of the
following three years by $100 (in her fourth year on college she withdraws $1500
each month). What must the monthly payment during the first three years be in
order that after Mary’s four years on college the balance on the savings account
is zero again, and what is the total interest Mary has earned after the seven
years?
6.9. Redo Example 6.11 for the case that the annual salary sm is paid during the
first nine months (monthly payment sm /9) in each year, i.e., there is no income
and hence no payment on the retirement savings account during the last three
months of the year. (This is the situation of university professors if they don’t
have additional income from grants.)
6.10. Assume the temperature of a roast in the oven increase at a rate proportional to
the difference between the oven (set to 400 degrees F) and the roast. If the roast
enters the oven at 50 degrees F and is measured one hour later to be at 90 when
134 Chapter 6 Modeling with Discrete Dynamical Systems

should the table be set if the eating temperature is 166 degrees F? Hint: write
down the difference equation and solve analytically.
6.11. Computer. This question concerns numerically exploring the solutions of the
equation
pn+1 = pn + αpn (1 − pn )
First determine all the equilibrium solutions of this difference equation by setting
p = pn+1 = pn . Now investigate the stability of these equilibrium numerically.
Consider the initial conditions
• p0 = 0
• p0 = 0.0001
• p0 = 2
Numerically simulate the difference equation using the following values of α
• α = .1
• α = .7
• α = 1.2
Describe your results and comment on the stability of the equilibrium you found.
Provide plots of all your results. It will make your comparisons easier if you plot
all the results for one value of α on a single graph.
6.12. Computer. This question concerns numerically exploring the solutions of the
equation
pn+1 = pn + 0.1pn (1 − pn )(2 − pn )
First determine all the equilibrium solutions of this difference equation. Numeri-
cally simulate the difference equation using the following initial conditions
• p0 = 0
• p0 = 0.0001
• p0 = .9999
• p0 = 1
• p0 = 1.0001
• p0 = 1.9999
• p0 = 2
• p0 = 2.0001
Describe your results and comment on the stability of the equilibrium you
found. Provide plots of all your results. It will make your comparisons easier if
you plot all the results on a single graph.
6.13. Computer. Simulate the fourth order difference equation

pn+4 = sin(pn+3 + pn+2 + pn+1 − pn ) + 2

and compare to the related equation

pn+4 = sin(pn+2 + pn+1 − pn ) + 2

using the initial coditions p1 = 6, p2 = 1, p3 = 2.5, p4 = −3. Explore mod-

ifications to these difference equations and see if you can find any interesting
behavior. For example, what is the effect of varying the nonhomogeneous term?
Plot your results in each case for 100 iterations.
 Section 6.5 Empirical Modeling 135

6.14. Computer. Consider the system of difference equations

xn+1 = 0.3xn + 0.8yn

yn+1 = 0.7xn + 0.2yn

Simulate these equations numerically for a variety of initial conditions and at-
tempt to determine any stable equilibrium. Verify that you have actually deter-
mined an equilibrium solution by substituting into the original system. (Note
that the equilibrium solution in this problem actually depends on the initial
condition.)
(a) How do the solutions change if you modify the first coefficient from 0.30 to
0.31?
(b) How do the solutions change if you modify the first coefficient from 0.30 to
0.31 and modify the 0.7 coefficient to 0.69.
(c) Compare the results in part a) and b). Can you explain?
6.15. Consider the difference equation

xn+2 + αxn+1 + βxn = 0

where it is assumed that α2 − 4β = 0. Show that xn = (− α2 )n n is a solution.

6.16. Find the linear second order nonhomogeneous difference equation relating the
price p2n+2 and p2n in the cobweb model. Solve this equation and produce a
convergence criterion. What does the equilibrium price converge to? Check your
result by computing the point of intersection of the supply and demand curves.
6.17. Determine analytical solutions to the following difference equations assuming in
each case that x1 = 1 and x0 = −1. Plot your results.
a) xn+2 + 3xn+1 + xn = 0
b) 10xn+2 + xn+1 + xn = 0
√
c) xn+2 + 3xn+1 + 43 xn = 0
6.18. Extend the population model with pairwise competition to include competition
among groups of three. Furthermore, assume that the competition among groups
of three is more intense than competition between pairs. Identity the new equi-
librium solution(s). Use a plot of ∆pn versus pn to argue whether the model
predicts a bounded population.
6.19. Consider a clam population that obeys the logistic difference Equation (6.26).
Modify this equation to account for constant harvesting of the clams. By com-
puting the new equilibrium points of the population model describe the impact
of harvesting on the clam population.
6.20. Consider three species A, B, C and the evolution of their populations an , bn and
cn .
• Species A eats B and C
• Species B eats neither A nor B
• Species C eats only A.
• Species B eats waste products produced by species A and B.
• The population of both species A and B increase in the absence of other
species.
• The population of species C decreases in the absence of A and B.
136 Chapter 6 Modeling with Discrete Dynamical Systems

• Species C is competes with itself for food while this is not true for species
A and B.
Write down a system of three coupled difference equations modeling the popu-
lations of the three species.
6.21. Computer. Provide a model for the bee colony population data in Table 6.2.
What does your model predict the long-term population to be?

day 1 2 3 4 5 6 7 8 9 10
number 20 25 60 85 111 146 177 182 184 171
day 11 12 13 14 15 16 17 18 19 20
number 179 167 161 146 159 154 162 166 166 168

TABLE 6.2: Bee colony population data.

6.22. Find the equilibria of the difference equation

pn+1 = pn − 0.1pn (1 − pn )(2 − pn )

and determine which of them are stable.

6.23. Computer. Numerically compute and plot 50 iterates of the difference equation

pn+1 = pn − 0.1pn (1 − pn )(2 − pn )

for each of the initial conditions

(a) (a) p0 = 0.9
(b) p0 = 1.1.
Is the behavior of the iterates consistent with the stability calculation of Problem
6.22?
6.24. Computer. Find all the equilibrium solutions of the logistic difference equation

xn+1 = rxn (1 − xn )

as a function of r. Letting x0 = 0.2 numerically iterate this difference equation

for 200 iterations for the following values of r:
• r=2
• r = 3.2
• r = 3.8282
• r = 3.83
Plot your results xn as a function of n for each case and comment. Does this
seem like a reasonable model for a population?
6.25. Consider the logistic difference equation with r > 0:

pn+1 = rpn (1 − pn ).

(a) Show that p1 = 0 is an equilibrium.

(b) Find the second equilibrium p2 (r). For which values of r is p2 (r) ≥ 0?
(c) For which values of r is p1 = 0 stable?
(d) For which values of r is p2 (r) stable?
 Section 6.5 Empirical Modeling 137

6.26. Computer. Numerically compute and plot 50 iterates of the difference equation

pn+1 = rpn (1 − pn )

for p0 = 0.5 and each of the following values of r:

(a) r = 0.8
(b) r = 2.9
(c) r = 3.1
(d) r = 3.5
(e) r = 3.9.
Describe the behavior of the iterates and relate it, where possible, to the stability
calculation of Problem 6.25.
6.27. Computer. Use a least squares approach to determine k in Newton’s Law of
cooling
Tn+1 = Tn + k(M − Tn )
using the data generated by our empirical fish model

Tn+1 = Tn + 0.01(M − Tn )1.25

First generate 200 points using this equation and compute k based on these
points. Now predict the next 200 points and calculate the error. If a fish is well
cooked at 170 degrees F how long does each model predict it will take to cook
the fish? Use the values M = 425 and T0 = 50.
6.28. Computer. Using the data provided in Table 6.3 estimate via least squares the
coeffients c1 , c2 , d1 , d2 in the model

an+1 = an + c1 an + d1 bn

bn+1 = bn + c2 an + d2 bn
Include your equations for the unknown coefficients in your write-up.

n an bn
1 15.00 45.00
2 30.00 30.00
3 24.00 36.00
4 26.40 33.60
5 25.44 34.56
6 25.82 34.18
7 25.67 34.33
8 25.73 34.27
9 25.71 34.29
10 25.72 34.28
11 25.71 34.29

TABLE 6.3: Did this data come from a linear system?

6.29. Extend the Equations (6.32) and (6.33) provided for computing the coefficients
c1 , c2 , g1 , g2 for the predator-prey model with no intra-species competition given
by Equation (6.27) to the case of Equations 6.29 where intraspecies competition is
accounted for. Your equations should now provide estimates for c1 , c2 , g1 , g2 , d1 , d2
138 Chapter 6 Modeling with Discrete Dynamical Systems

6.30. Consider the differential equation for the unforced damped nonlinear pendulum
d2 x dx
+α + sin x = 0
dt2 dt
where x(t) represents the angular displacement from the equilibrium in radians.
Using the expressions for the numerical estimates of the derivatives
d2 x x + xn−1 − 2xn
= n+1
dt2 (∆t)2
and
dx xn − xn−1
=
dt ∆t
where xn ≡ x(n∆t).
(a) Show that the differential equation can be approximated by the second order
difference equation
xn+1 = (2 − α∆t)xn + (α∆t − 1)xn−1 − (∆t)2 sin(xn ) (6.34)
(b) Simulate this difference equation for 1000 iterations using the values ∆t =
0.05, α = 0.1, x1 = 0, x2 = 0.0001 and plot your result. Repeat this calcula-
tion for x1 = 2, x2 = 2.0001 and compare your results.
(c) Redo this simulation using the small angle approximation sin x = x, i.e.,
simulate
xn+1 = (2 − α∆t)xn + (α∆t − 1)xn−1 − (∆t)2 xn (6.35)
using the values ∆t = 0.05, α = 0.1, x1 = 0, x2 = 0.0001 and plot your
result. Again, repeat this calculation for x1 = 2, x2 = 2.0001 and compare
your results with those found in part (c).
(d) Rewrite the second order Equation (6.34) as a system of two first order
equations via the substitution yn+1 = xn and determine all equilibria. Note
that the equilibria can also be determined directly from Equation (6.34).
(e) By computing the eigenvalues of the Jacobian matrix of this system, ascer-
tain which equilibria are stable and unstable. Discuss.
6.31. Repeat parts (d) and (e) of Problem 6.30 for the small angle approximation
Equation (6.35) and compare.
6.32. Analytically solve the linear difference equation from the previous problem
xn+1 = (2 − α∆t)xn + (α∆t − 1)xn−1 − (∆t)2 xn
and compare with your numerical simulation above. For simplicity you may take
∆t = 0.05, α = 0.1, x1 = 2, x2 = 2.0001.
6.33. Analytically solve the linear nonhomogeneous difference equation
xn+1 = (2 − α∆t)xn + (α∆t − 1)xn−1 − (∆t)2 xn + 0.01 sin(n/50)
Simulate this problem numerically and compare with your analytical solution for
2000 iterations. Can you identify a transient (i.e., a term that goes to zero) and
steady state (persistent) components of your solution? Again, for simplicity you
may take ∆t = 0.05, α = 0.1, x1 = 2, x2 = 2.0001. Hint: combine your solution
to the homogeneous problem found above with a particular solution of the form
pn = A cos(n/50) + B sin(n/50)
Solve for the undetermined coefficients A and B .
 Section 6.5 Empirical Modeling 139

REFERENCES
[1] Bagnet, G. C., 2206, The widget maker’s guide to snarfle splatting and freen
wongling, 17th Edition, Buena Free Press, Crawdadsville, South Vermont.

[2] D. Knuth, Notices Amer. Math. Soc. 49 (2002), no. 3, 318–324.

[3] I. Lepper, Theoret. Comput. Sci. 269 (2001), no. 1-2, 433–450.

[4] R. R. Fletcher, III, Congr. Numer. 147 (2000), 17–31.