1

Discrete Difference Equations

from Mathematics for the Life Sciences, book by Bodine, Lenhart and Gross

These notes discuss how we can use sequences from data or from theory to determine how
biologically relevant variables change. One of the main objectives is to learn how to use mathe-
matical formulae to describe a sequence and what the long-term implications are for the variable
being measured. One example is using your sequence of heights measured from birth to age 10
to predict what your adult height would be. Note that we are using height as an example rather
than weight because adult height is much more constrained, and does not change much over
adulthood, as compared to body weight (you might consider why this is so).

EXAMPLE: Bird Count

Every Christmas the Audubon Society invites birders across the United States to participate
in the Christmas Bird Count. Data for various birds have been collected for over 100 years. The
data is posted on the Audubon Society’s website. Suppose we form a sequence using the data
collected on red cardinals each year. The count the first year would be a1 , the second year would
be a2 , the third year would be a3 , and so on.

Year Count
Year Count Year Count Year Count
1998 5439
1959 2206 1972 3696 1985 5359
1999 4367
1960 2297 1973 4989 1986 4321
2000 6045
1961 2650 1974 3779 1987 5044
2001 4632
1962 2277 1975 4552 1988 3092
2002 6974
1963 2242 1976 3872 1989 5388
2003 4528
1964 2213 1977 4049 1990 4079
2004 6875
1965 2567 1978 4037 1991 4416
2005 5154
1966 3152 1979 3475 1992 4828
2006 6631
1967 2186 1980 4448 1993 4291
2007 7051
1968 2998 1981 3660 1994 4861
2008 4882
1969 2628 1982 5141 1995 4662
2009 6896
1970 3450 1983 4890 1996 4827
2010 6190
1971 2829 1984 3500 1997 4377
2011 6739

Table 1: Data collected by Tennessee birders for the Audubon Christmas Bird Count. Data
reflects the total count of Northern Cardinals sighted in Tennessee.

Cardinals: Birders in Tennessee have reported seeing the Northern Cardinal (Cardinalis
cardinalis) since 1959. The numbers reported are given in the Table 1. We could make a
sequence out of this data. If an represents the count at year n, where n = year − 1959 then,
a0 = 2206, a1 = 2297, a3 = 2650, . . . , a50 = 6896, a51 = 6190, a52 = 6739.
This particular sequence has 53 terms in the sequence.
A real sequence is a function f : N → R with domain being the natural numbers N and the
range being contained in the real numbers R. We write an = f (n) and call an the nth term in
the sequence.
 2

Some real sequences can be expressed by an algebraic formulation.

EXAMPLE: Sequence
2n
Suppose we defined the sequence an = f (n) = (−1)n n+1 . Find the first 5 terms of this
sequence.
Solution: We can use the function definition of the sequence to find each term.

2·1
f (1) = (−1) = −1
1+1
2·2 4
f (2) = (−1)2 =
2+1 3
2·3 6 3
f (3) = (−1)3 =− =−
3+1 4 2
2·4 8
f (4) = (−1)4 =
4+1 5
2·5 10 5
f (5) = (−1)5 =− =−
5+1 6 3

Limit of a Sequence
Our objective is to point out how we can use mathematical descriptions of sequences to
analyze the behavior of simple models.
Suppose we have a sequence {an } whose terms are getting closer and closer to L as n gets
larger and larger. Then we say that the limit of the sequence {an } is L. Formally, we write

lim an = L
n→∞

if the the sequence {an } converges to the limit L (a finite number). If the sequence {an } increases
toward infinity or decreases toward negative infinity as n → ∞, then we say that the sequence
{an } does not have a limit. Thus, note that not all sequences have limits.

Discrete Difference Equations

Suppose we have a population that doubles each year. We could construct a sequence, where
xn represents the size of the population at each time step. Since the population is doubling each
year, xn+1 = 2xn , i.e. the population at time step n + 1 is twice what it was at time step n,
using yearly time steps.
We will study equations showing biological quantities that changes over time. For example,
the Fibonacci sequence,
1, 1, 2, 3, 5, 8 . . .
(originally involving the births in generations of rabbit populations), is given by:

x0 = 1, x1 = 1
 3

and then for n = 1, 2, . . .

xn+1 = xn + xn−1 .
Note that xn+1 is a function of the two previous terms in the sequence, which means that this
equation for xn+1 is called a second order difference equation. Note that the equation above,
xn+1 = 2xn , is called a first order difference equation since each new term in the sequence only
depends on the previous term.
If the xn+1 term can be written solely as a function of the xn term, i.e.

xn+1 = f (xn ),

then we say that this sequence is built from a first order difference equation or recurcive
equation. The term ‘first order’ meaning that to find xn+1 , you only need to use the previous
value, xn . Notice, that if we know x0 (the initial value in the sequence) then we can determine
all the terms in the sequence, coming from a first difference equation.

EXAMPLE: Doves
A population of doves increases by 3% each year. Let xn be the size of the population at year
n. Then xn+1 = xn + .03xn = 1.03xn . Thus, the first order difference equation that describes
the population is
xn+1 = 1.03xn .
Notice that if we know x0 , then

x1 = 1.03x0
x2 = 1.03x1 = 1.03(1.03x0 ) = (1.03)2 x0
..
.
xn = (1.03)n x0 .

Geometric and Arithmetic Sequences

The sequence in the dove example above is called a geometric sequence since its population
changes by a multiplication factor each time step.
A geometric sequence is defined by

xn+1 = rxn

where r is a fixed real number. Notice, if we know x0 , then

x1 = rx0
x2 = rx1 = r(rx0 ) = r2 x0
x3 = rx2 = r(r2 x0 ) = r3 x0
..
.
xn = r n x0
 4

The general solution to difference equation represents the xn in terms of x0 , n and other given
constants. For a geometric sequence, the general solution is is

xn = r n x0 .

When 0 < r < 1, the sequence decays to zero. It has a limit of zero, meaning as n get large
(n → ∞), xn → 0. For r > 1, the terms of the sequence increase exponentially.

EXAMPLE: Hares
A population of wild hares increases by 13% each year. Currently, there are 200 hares. If xn
is the number of hares in the population at the end of year n, find

1. the difference equation relating xn+1 to xn ,

2. the general solution to the difference equation found in (a), and

3. the number of hares in the population at the end of six years from now.

Solution: Notice that x0 = 200.

1. Since the population increases by 13% each year,

xn+1 = xn + 0.13xn = 1.13xn .

|{z} | {z }
from year before increase

2. Using the notation of the definition of a geometric sequence, here r = 1.13, thus the general
solution to the difference equation in (a) is

xn = (1.13)n x0 = 200(1.13)n .

3. For n = 6, x6 = 200(1.13)6 ≈ 416. Thus, at the end of year six there are approximately
416 hares.

If a population increases by a fixed number d each time period, we say the sequence is an
arithmetic sequence,
xn+1 = xn + d.
Suppose we know the intial value x0 , and then the general solution to an arithmetic sequence
is:

x1 = x0 + d
x2 = x1 + d = (x0 + d) + d = x0 + 2d
..
.
xn = x0 + nd.
 5

Linear Difference Equation with Constant Coefficients

A first order difference equation is linear if it takes the form
xn+1 = an xn + bn
where an and bn are sequences of constants. In this section, we are interested in linear first order
difference equations where an = a and bn = b for all n. That is, we can write the difference
equation as
xn+1 = axn + b.
Let us find the general solution to this difference equation.
First, notice that if a = 1, then we are in the case of the arithmetic sequence and we have
already found the general solution,
xn = x0 + nb.
Second, if b = 0, we have a geometric sequence with the general solution xn = an x0 .
If a 6= 1 and b 6= 0, then to find the general solutions, we use the following steps.

Finding the General Solution to xn+1 = axn + b where a 6= 1

Step 1: We first solve the homogeneous difference equation, which is the difference equation
without the constant b, that is xn+1 = axn . We already know the solution of this has the form
an c where c is some constant. Notice, we cannot say that c = x0 here because a solution to the
homogeneous equation is not necessarily a solution to the full equation.
Step 2: Next, we construct a particular solution pn . We will assume the particular solution
is a constant, that is pn = K for all n. We now need to determine K. For pn to be a solution
to the difference equation, it must satisfy
pn+1 = apn + b.
However, pn = K, thus
K = aK + b
K − aK = b
(1 − a)K = b
b
K = .
1−a

b
Thus, our particular solution is pn = 1−a .
Step 3: Lastly, we form the general solution as the sum of the homogeneous and particular
solutions. That is,
b
xn = can + .
1−a
b b
Notice x0 = c + 1−a , thus c = x0 − 1−a .
To see why building a general solution in this way works, consider two solutions, sn and wn ,
to
xn+1 = axn + b.
 6

Let yn = sn − wn and then yn satisfies this difference equation,

yn+1 = sn+1 − qn+1 = (asn + b) − (awn + b) = a(sn − wn ) = ayn .

We have yn solving the homogeneous equation, giving yn = can for a constant c. Any general
solution can be written as sn = can + wn .
Therefore, the general solution to the difference equation is
 
b b
xn = x0 − an + .
1−a 1−a

Let us consider an example of a population that could be modeled by a linear difference

equation with constant coefficients. What would each term in the difference equation represent?

xn+1 = axn − b
|{z} (1)
|{z} |{z}
pop. @ t = n + 1 pop growth or decline fixed decrease

Notice that in Equation (1), the population grows or declines first (with the factor a) and
then a fixed amount b is being removed. Thus, this model could represent a population that is
being harvested by a fixed amount at each time step after growth or decline. If the constant b
was being added in the equation, then the difference equation might model a population (after
growth or decline) that was being augmented by a fixed amount each time step (think of a fish
population being restocked each season).
About the order of events: If one harvests the population before the growth happens, then
you would start with this equation:

xn+1 = a(xn − b).

Furthermore, it may be advantageous to think of the constant a, representing population

growth or decline, as the difference between the birth and death rates of the population. That
is, if β is the proportion by which the population increases due to births at each time step, and
δ is the proportion by which the population decreases due to deaths at each time step, then
a = 1 + β − δ. If a > 1 then the population is growing (in the absence of harvesting), if a < 1
then the population is decreasing (in the absence of harvesting), and if a = 1 then in the absence
of harvesting the population will remain constant.

EXAMPLE: Fisheries
Consider a lake fish population whose yearly birth rate is 1.2, and yearly death rate is 0.7.
Each year, fishing is allowed until 1200 fish are caught. Thereafter, fishing is banned. Currently,
there are 12,230 fish in the lake.
1. Write a difference equation for the lake fish population and find the general solution.

2. How many fish are in the lake after 5 years?

3. If the resource managers of the lake wanted the population to remain constant each year,
what level of harvesting should they allow?
 7

Solution: Let xn be the size of the fish population at the end of year n. Then x0 = 12, 230.

1. Since β = 1.2 and δ = 0.7,then a = 1 + β − δ = 1 + 1.2 − 0.7 = 1.5. Thus, xn+1 =

1.5xn − 1200. The general solution is then
 
−1200 −1200
xn = x0 − (1.5)n +
1 − 1.5 1 − 1.5

= (12, 230 − 2400)(1.5)n + 2400

= 9830(1.5)n + 2400.

2. For n = 5, xn = 9830(1.5)5 + 2400 ≈ 77, 047. Thus, there are approximately 77,047 fish
in the lake after 5 years.

3. To keep the population constant, we would want xn+1 = xn for all n. Specifically x1 = x0 .
Let h be the harvesting level we wish to find. Thus, using x1 = x0 in

x1 = 1.5x0 − h,

we want to solve for h.

x0 = 1.5x0 − h
(1 − 1.5)x0 = −h
−0.5x0 = −h
−0.5(12, 230) = −h
h = 6115

Thus, if the lake resource managers allowed 6115 fish each season to be caught, the fish
population size will remain constant from year to year.

EXAMPLE: Stocking a Lake

Consider a lake fish population whose yearly birth rate is 0.5, and yearly death rate is 0.7.
Currently, there are 12,230 fish in the lake. How many fish are needed to stock the lake each
year, so that the population remains constant?
Solution: The birth and death rates correspond to β = 0.5 and δ = 0.7, and thus a =
1 + β − δ = 1 + 0.5 − 0.7 = 0.8. Thus, the population is naturally decreasing. We can construct
the difference equation
xn+1 = 0.8xn + α
where xn is the size of the fish population at the end of year n, and α is the number of fish
that are added to the lake each year. Notice that x0 = 12, 230. To find the number of fish
needed to stock the lake each year to keep the population constant, we use x1 = x0 and solve
x1 = 0.8x0 + α for α.

x0 = 0.8x0 + α
 8

(1 − 0.8)x0 = α
0.2x0 = α
α = 0.2(12, 230) = 2446

Thus, each year the lake must be stocked with an additional 2,446 fish to maintain a constant
population size.

Data: Log-Log & Semi-Log Graphs

Data can be displayed with different types of scales on the horinzontal and vertical axes.
Frequently in plots of biological data, the horizontal axis may be labeled

ln(x)

and the vertical axis may be labeled

ln(y),
where x and y are the biological variables under consideration. Such a graph is called a log-log
graph. If the horizontal axis is labeled
x
and the vertical axis may be labeled
ln(y),
where x and y are the biological variables under consideration, then the graph is called semi-log.
Examples of semi-log graphs can be found in the next two examples.
We will see how the log-log and semi-log graphs relate to allometric and expontial functions.
We will see that if our data is close to an allometric (power) or exponential function, we can
rescale the our data and see that the transformed data will look like a line on a log-log or
semi-log graph.
Next, consider the exponential function f (x) = acx where a and c are constants. Let us take
the natural log of both sides of the equation and see what happens.

ln (f (x)) = ln (acx )
= ln a + ln cx
= ln a + x ln c

Does this look like the equation of a line? Look closer using y = ln (f (x)),

constant constant
z}|{ z}|{
ln y = ln a + |{z}
x ln c .
|{z}
variable variable

We see that the y-intercept is ln a and the slope is ln c (if c = e ≈ 2.71828... then ln c = 1).
Notice the original vertical axis variable f (x) has been rescaled to ln (f (x)). Notice also that
we could have just as easily used log instead of ln.
 9

Now, consider the allometric (power) function g(x) = axc where a and c are constants.
Again, let us take the natural log of both sides of the equation and see what happens.

ln (g(x)) = ln (axc )
= ln a + ln (xc )
= ln a + c ln x

Again, if we take a closer look we should see the equation of a line using y = (g(x)),

constant
constant
z}|{ z}|{
ln y = ln a + c ln x
|{z}
|{z}
variable variable

where the y-intercept is ln a and the slope is c. Here, in the allometric case, the vertical axis
variable g(x) has been rescaled to ln (g(x)) and the horizontal axis variable has been rescaled
to ln x. Notice this is different from the exponential case in which only the vertical axis was
rescaled.
What is the big picture?

? Given an exponential equation, if we rescale the y-axis to be logarithmic, we get an equa-

tion for a line.

? Given an allometric (power) equation, if we rescale both the x and y axes to be logarithmic,
we get an equation for a line.

Linking Rescaling with Linear Regressions

Suppose you were given a set of data that appeared to have an exponential or allometric
trend. How would you determine the exponential or allometric function that best describes that
set of data? If the data had a linear trend then we could use linear regression to find the best fit
line for the data. We have just seen how we can transform exponential and allometric equations
to linear equations with rescaling. Using this same idea, we can rescale data so that it appears
to be linear so that we may then use the techniques of linear regression to find the equation that
best describes the data.
EXAMPLE:Harvesting Bluefish
For the first example, we consider some data on Bluefish harvesting in the Chesapeake Bay,
reported by the Daily Press, a newspaper in Virginia.
 10

Year Bluefish (lbs)

1940 15,000
1945 150,000
1950 250,000
1955 275,000
1960 270,000
1965 280,000
1970 290,000
1975 650,000
1980 1,200,000
1985 1,500,000
1990 2,750,000

Hopefully, looking at the data plotted out, the first observation you make is that the data are
not linearly related. Let us try rescaling the y-axis data on a logarithmic scale. Additionally,
on the horizontal axis, let x = 15 (year − 1940).

x y ln y
Year Bluefish (lbs) Rescaled
0 15,000 9.616
1 150,000 11.918
2 250,000 12.429
3 275,000 12.525
4 270,000 12.506
5 280,000 12.543
6 290,000 12.578
7 650,000 13.385
8 1,200,000 13.998
9 1,500,000 14.221
10 2,750,000 14.827

With the exception of the data point corresponding to the year 1940, the data appear to be
more linearly related when the y-axis data is rescaled. If we fit a least-squares regression line to
this data, we find the equation for the best fit line is

ln y = 0.3797x + 10.8784

where x is the year and y is the pounds of bluefish harvested. We can convert this equation to
an exponential equation:

ln y = 0.3797x + 10.8784
eln y = e0.3797x+10.8784
y = e0.3797x × e|10.8784
{z }
constant
 11

y = (53, 019) e0.3797x .

Using this equation we could extrapolate how many fish were caught in 2000. First, we compute

1
x= (2000 − 1940) = 12.
5

Then,
y = (53, 019) e0.3797(12) = 5, 049, 529.
Thus, we estimate that in the year 2000, about 5.05 million pounds of blue fish were caught.
EXAMPLE: Mutation Rates
Researchers studying the relationship between the generation time of a species and the
mutation rate for genes that cause deleterious effects gathered the following data.

Generation Genomic
Species Time Mutation Rate
(in years) (per generation)
D. melanogaster/D. pseudoobscura 0.1 0.070
D. melanogaster/D. simulans 0.1 0.058
D. picticornis/D. silvestris 0.2 0.071
Mouse/rat 0.5 0.50
Chicken/old world quail 2 0.49
Dog/cat 4 1.6
Sheep/cow 6 0.90
Macaque/New World Monkey 11 1.9
Human/chimpanzee 25 3.0

Let x = the generation time and y = the genomic mutation rate. If we rescale both the
vertical and the horizontal axes then we have the following data
 12

ln x ln y
-2.3 -2.659
-2.3 -2.847
-1.6 -2.645
-0.7 -0.693
0.7 -0.713
1.4 0.470
1.8 -0.105
2.4 0.642
3.2 1.099

If we fit a least-squares regression line to this data, we find the equation for the best fit line
is
ln y = 0.7097 ln x − 1.0316.
We can convert this equation to an allometic (power function) equation using the properties of
logarithms and exponentials,

ln y = ln x0.7097 − 1.0316
0.7097 −1.0316
eln y = eln x
0.7097 −1.0316
eln y = eln x e
0.7097 −1.0316
y = x e
y = 0.3564x0.7097 .

Using this equation, if we knew a certain species had a generation time of 5 years, we could
interpolate the genomic mutation rate of this species,

y = 0.3564(10)0.7097 = 0.1827.

Thus, we would estimate that this particular species has a genomic mutation rate of 0.1827
mutations per generation.

Exercise 1.
Warm blooded animals use large quantities of energy to maintain body temperature because
of heat loss through body surface. In fact, biologists believe that the primary energy drain on a
resting warm blooded animal is the maintenance of body temperature. The table below shows
the body weights in grams (g) and pulse rates in beats per minute (bpm) for nine different bird
species.
 13

Bird Body Weight (g), x Pulse Rate (bpm), y

Canary 20 1000
Pigeon 300 185
Crow 341 378
Buzzard 658 300
Duck 1100 190
Hen 2000 312
Goose 2300 240
Turkey 8750 193
Ostrich 71,000 65

(a) Make a hypothesis about how you think body weight and pulse rate are related in birds?
How might your hypothesis explain birds maintaining their body heat.

(b) What type of function (linear, exponential, or allometric (power)) best describes the rela-
tionship between the weight and pulse rate data? Explain. Find the equation for the the
function that best describes the data as y in terms of x.

Exercise 2.
Researchers measured the diameters of 20 trees in a central Amazon rain forest and used
14
C-dating to determine the ages of these trees. The data are given in the following table.

Diameter (cm), x Age (yr), y Diameter (cm), x Age (yr), y

180 1372 115 512
120 1167 140 512
100 895 180 455
225 842 112 352
140 772 100 352
142 657 118 249
139 582 82 249
150 562 130 227
110 562 97 227
150 552 110 172

Consider the use of diameter x, as a predictor of age y.

(a) Make a scatter plot of age on the vertical axis and diameter on the horizontal axis/

(b) What type of function (linear, exponential, or allometric) best describes the relationship
between the age and diameter data? Explain. Write the function for y in terms of x.

Exercise 3.
In an attempt to measure how the pace of city life is related to the size of the city, two
researchers measured the mean speed of pedestrians in 15 cities by measuring the mean time it
took them to walk 50 feet.
 14

City Population (x) Speed (ft/s) (y)

Brno, Czechoslovakia 341,948 4.81
Prague, Czechoslovakia 1,092,759 5.88
Corte, France 5,491 3.31
Bastia, France 49,375 4.90
Munich, Germany 1,340,000 5.62
Psychro, Crete 365 2.67
Itea, Greece 2,500 2.27
Iráklion, Greece 78,200 3.85
Athens, Greece 867,023 5.21
Safed, Israel 14,000 3.70
Dimona, Israel 23,700 3.27
Netanya, Israel 70,700 4.31
Jerusalem, Israel 304,500 4.42
New Haven, CT, USA 138,000 4.39
Brooklyn, NY, USA 2,602,000 5.05

(a) Plot the original pairs of numbers (x, y). Is the pattern linear or nonlinear?

(b) Compute R2 , for the data.

(c) Plot y against ln x. Are the data more linear now than in part (a)?

(d) Compute R2 for y against ln x. Is R2 closer to 1 than in part (b)? What does this say about
how the data are related?

(e) Find the equation for the least squares regression line of y against ln x.

Partial answers

Exercise 1
(a) As weight increases, then pulse decreases. Pulse rates can affect the maintaining of body
heat.
(b) allometric, ln(y) = −0.28 ln x + 7.5 and y = x0.28 e7.5

Exercise 2
allometric, y = 1.24x1.23 .

Exercise 3
(a) Nonlinear,
(b) R2 = 0.584,
(c) Yes, the data looks more linear,
(d) R2 = 0.845 which is close to 1 which means the original data are not linearly related,
(e) y = 0.410 ln x − 0.307.