197

“Words differently arranged have a different meaning and meanings differently arranged
have a different effect.”
Blaise Pascal (1623-1662)

Chapter 6

Difference Equations and Z-transforms

There are many concepts in science and engineering that can be approached
from either a discrete or a continuous viewpoint. For example, consider how you
might record the temperature outside at some speciÞc place as a function of time.
One technique would be to purchase a chart recorder capable of measuring and
plotting the temperature as a function of time. This would give a continuous
record of the temperature over some interval of time. Another way to record
the temperature would be to measure the temperature, at the speciÞed place,
at discrete time intervals. The contrast between these two methods is that one
method measures temperature continuously while the other method measures
the temperature in a discrete fashion.
In any laboratory experiment, one must make a decision as to how data from
the experiment is to be collected. Whether discrete measurements or continuous
measurements are recorded depends upon many factors as well as the type of
experiment being considered. The techniques used to analyze the data collected
depends upon whether the data is continuous or discrete.
The investment of money at compound interest is an example of a physical
problem which requires analysis of discrete values. Say, $1,000.00 is to be in-
vested at R percent interest compounded quarterly. How do we determine the
discrete values representing the amount of money available at the end of each
compound period? To solve this problem, we let P0 denote the amount of money
R
initially invested, R the percent interest yearly with 14 100 = i the quarterly inter-
est and Pn the principal due at the end of the nth compound period. We can
198
then set up the equations for the determination of Pn . We have
P0 = Initial amount invested
P1 = P0 + P0 i = P0 (1 + i)
P2 = P1 + P1 i = P1 (1 + i) = P0 (1 + i)2
P3 = P2 + P2 i = P2 (1 + i) = P0 (1 + i)3
..
.
Pn = Pn−1 + Pn−1 i = Pn−1 (1 + i) = P0 (1 + i)n

R
For i = 14 100 and P0 = 1, 000.00, Þgure 6-1 illustrates a graph of Pn vs time, for
a 30 year period, where one year represents four payment periods. In this Þgure
values of R for 4%, 5.5%, 7%, 8.5% and 10% were used in the above calculations.

Figure 6-1. Return from $1,000 investment compounded quarterly over 30 year period.
In this chapter we investigate some techniques that can be used in the analysis
of discrete phenomena like the compound interest problem just considered.
The study of calculus has demonstrated that derivatives are the mathemat-
ical quantities that represent continuous change. We Þnd that if we replace
derivatives (continuous change) by differences (discrete change), then linear or-
dinary differential equations become linear difference equations. We shall inves-
tigate these difference equations and Þnd ways to construct solutions to such
equations.
In the following discussions, note that the various techniques developed for
analyzing discrete systems are very similar to many of the methods used for
studying continuous systems.
 199
Differences and Difference Equations

Consider the function y = f (x) illustrated in the Þgure 6-1 which is evaluated
at the equally spaced x−values of x0 , x1 , x2 , . . . , xn , xn+1 , . . . where xn+1 = xn + h for
n = 0, 1, 2, . . . , where h is the distance between two consecutive points.

Figure 6.2. Discrete values of y = f (x).

dy
Let yn = f (xn ) and consider the approximation of the derivative dx at the
discrete value xn . By using the deÞnition of a derivative we may write the
approximation as ! dy ! yn+1 − yn
! ≈ .
dx x=xn h
This is called a forward difference approximation. By letting h = 1 in the above
equation, we can deÞne the Þrst forward difference of yn as

∆yn = yn+1 − yn . (6.1)

There is no loss in generality in letting h = 1, since we can always rescale the

x-axis by deÞning the new variable X deÞned by the transformation equation
x = x0 + Xh, then when x = x0 , x0 + h, x0 + 2h, . . . , x0 + nh, . . . the scaled variable X
takes on the values X = 0, 1, 2, . . . , n, . . . .
DeÞne the second forward difference as a difference of the Þrst forward dif-
ference. A second difference is denoted by the notation ∆2 yn and
∆2 yn = ∆(∆yn ) = ∆yn+1 − ∆yn = (yn+2 − yn+1 ) − (yn+1 − yn )
(6.2)
or ∆2 yn = yn+2 − 2yn+1 + yn .
Higher ordered difference are deÞned in a similar manner. A nth order forward
difference is deÞned as differences of (n − 1)st forward differences for n = 2, 3, . . . .
200
d
Analogous to the differential operator D = dx , there is a stepping operator E
deÞned as follows:
Eyn = yn+1
E 2 yn = yn+2
(6.3)
···
E m yn = yn+m .
From the deÞnition given by equation (6.1) we can write the Þrst ordered differ-
ence
∆yn = yn+1 − yn = Eyn − yn = (E − 1)yn

which illustrates that the difference operator ∆ can be expressed in terms of the
stepping operator E and
∆ = E − 1. (6.4)

This operator identity, enables us to express the second-order difference of yn as

∆2 yn = (E − 1)2 yn
= (E 2 − 2E + 1)yn
= E 2 yn − 2Eyn + yn
= yn+2 − 2yn+1 + yn .

Higher order differences such as ∆3 yn = (E −1)3 yn , ∆4 yn = (E −1)4 yn , . . . and higher

ordered differences are quickly calculated by applying the binomial expansion to
the operators operating on yn .
Difference equations are equations which involve differences. For example,
the equation
L2 (yn ) = ∆2 yn = 0

is an example of a second-order difference equation, and

L1 (yn ) = ∆yn − 3yn = 0

is an example of a Þrst-order difference equation. The symbols L1 (), L2 () are

operator symbols. Using the operator E , the above equations can be written as
L2 (yn ) = ∆2 yn = (E − 1)2 yn = yn+2 − 2yn+1 + yn = 0 and
L1 (yn ) = ∆yn − 3yn = (E − 1)yn − 3yn = yn+1 − 4yn = 0,

respectively.
 201
There are many instances where variable quantities are assigned values at
uniformly spaced time intervals. We shall be interested in studying these discrete
variable quantities by using differences and difference equations. An equation
which relates values of a function y and one or more of its differences is called
a difference equation. In dealing with difference equations one assume that the
function y evaluated at xn and its differences ∆yn , ∆2 yn , . . ., are all deÞned for
every number x in some set of values {x0 , x0 +h, x0 +2h, . . . , x0 +nh, . . .}. A difference
equation is called linear and of order m if it can be written in the form
L(yn ) = a0 (n)yn+m + a1 (n)yn+m−1 + · · · + am−1 (n)yn+1 + am (n)yn = g(n), (6.5)

where the coefficients ai (n), i = 0, 1, 2, . . . , m, and the right-hand side g(n) are
known functions of n. If g(n) #= 0, the difference equation is said to be nonhomo-
geneous and if g(n) = 0, the difference equation is called homogeneous.
The difference equation (6.5) can be written in the operator form
L(yn ) = [a0 (n)E m + a1 (n)E m−1 + · · · + am−1 (n)E + am (n)]yn = g(n),

where E is the stepping operator.

A m th-order linear initial value problem associated with a m th-order linear
difference equation consists of a linear difference equation of the form given in
the equation (6.5) together with a set of m initial values of the type
y0 = α0 , y1 = α1 , y2 = α2 , ..., ym−1 = αm−1 ,

where α0 , α1 , . . . , αm−1 are speciÞed constants.

Difference equations may be solved using techniques which are very similar
to the solution methods associated with ordinary differential equations. All the
concepts and theorems derived for linear differential equations have analogs in
the study of linear difference equations. Instead of presenting each derivation,
we list the following summary of these important results. In this summary D.E.
can represent either “differential equation” or “difference equation”.

1. An nth-order, linear, homogeneous D.E. possesses n independent solutions.

2. The general solution of an nth-order, linear D.E. has n arbitrary constants.
3. The general solution of an nth-order, linear, nonhomogeneous D.E. can be
formed by adding the general complementary solution of the homogeneous
equation to any particular solution of the nonhomogeneous equation.
202
4. If two independent solutions of a linear, homogeneous D.E. are known, then
any linear combination of these solutions is also a solution.
5. If n-independent solutions to a linear, nth-order, homogeneous D.E. are
known, then any linear combination of these solutions produces the general
solution.
6. An nth-order linear initial value problem associated with a D.E. possesses a
unique solution.
The above analogies between difference equations and differential equations
can be anticipated if one writes the forward difference approximation of a deriva-
tive in operator form as Dy = limh→0 ∆y d
h , where D = dx and ∆ is the difference
operator.

Example 6-1.
Show ∆ak = (a − 1)ak , for a constant and k an integer.
Solution: Let yk = ak , then by deÞnition

∆yk = yk+1 − yk = ak+1 − ak = (a − 1)ak .

Example 6-2.
The function
k [N] = k(k − 1)(k − 2) · · · [k − (N − 2)][k − (N − 1)], k [0] ≡ 1

is called a factorial polynomial, see equation (4.72). Here k[N] is a product of N

terms.
Show ∆k[N ] = N k[N−1] for N a positive integer and Þxed.
Solution: Observe that the factorial polynomials are

k [0] = 1, k [1] = k, k[2] = k(k − 1), k[3] = k(k − 1)(k − 2), ···

Use yk = k[N ] and calculate the forward difference

∆yk = yk+1 − yk = (k + 1)[N] − k [N]
= (k + 1) (k)(k − 1) · · · [k + 1 − (N − 1)] − k(k − 1)(k − 2) · · · [k − (N − 2)][k − (N − 1)]
" #$ % " #$ %
k[N −1] k [N−1]

which simpliÞes to
∆yk = {(k + 1) − [k − (N − 1)]} k[N −1] = N k [N −1] .
 203
Example 6-3.
Verify the forward difference relation

∆(Uk Vk ) = Uk ∆Vk + Vk+1 ∆Uk

Solution: Let yk = Uk Vk , then we can write

∆yk = yk+1 − yk
= Uk+1 Vk+1 − Uk Vk + [Uk Vk+1 − Uk Vk+1 ]
= Uk [Vk+1 − Vk ] + Vk+1 [Uk+1 − Uk ]
= Uk ∆Vk + Vk+1 ∆Uk .

Special Differences

The table 6.1 contains a list of some well known forward differences which
are useful in many applications. The veriÞcation of these differences is left as an
exercise.

Table 6.1 Some common forward differences

1. ∆ak = (a − 1)ak
k[N] is factorial function
2. ∆k[N ] = N k [N−1] N Þxed See equation (4.72)
3. ∆ sin(α + βk) = 2 sin(β/2) cos(α + β/2 + βk) α, β constants
4. ∆ cos(α + βk) = −2 sin(β/2) sin(α + β/2 + βk) α, β constants
& ' & '
k k (k)
5. ∆ = N Þxed N are binomial coefficients
N N −1
6. ∆(k!) = k(k!)
7. ∆(Uk Vk ) = Uk ∆Vk + Vk+1 ∆Uk
& '
1 −N k[N] is factorial function
8. ∆ = , N Þxed
k[N ] k[N +1] See equation (4.72)
9. ∆k 2 = 2k + 1
10. ∆ log k = log (1 + 1/k)
204
Finite Integrals
Associated with Þnite differences are Þnite integrals. If ∆yk = fk , then the
function yk , whose difference is fk , is called the Þnite integral of fk . The inverse
of the difference operation ∆ is denoted ∆−1 and one can write yk = ∆−1 fk ,
if ∆yk = fk . For example, consider the difference of the factorial function k[n] ,
deÞned by equation (4.72). If ∆k[n] = nk[n−1] , then ∆−1 nk[n−1] = k[n] . Associated
with the difference table 6.1 is the Þnite integral table 6.2. The derivation of the
entries is left as an exercise.
Table 6.2 Some selected Þnite integrals
ak
1. ∆−1 ak = a #= 1
a−1
k [n+1] k[n] is factorial function
2. ∆−1 k[n]) = See equation (4.72)
n+1
−1
3. ∆−1 sin(α + βk) = cos(α − β/2 + βk) α, β constants
2 sin(β/2)
1
4. ∆−1 cos(α + βk) = sin(α − β/2 + βk) α, β constants
2 sin(β/2)
& ' & '
−1 k k (k )
5. ∆ = nf ixed n are binomial coefficients
n n+1
(a + bk)[n+1]
6. ∆−1 (a + bk)[n] = a, b constants.
b(n + 1)

Summation of Series
Let yk+1 − yk = fk , then one can substitute k = 0, 1, 2, . . . to obtain
y1 − y0 =f0
y2 − y1 =f1
y3 − y2 =f2
.. (6.6)
.
yn − yn−1 =fn−1
yn+1 − yn =fn

Adding these equations one obtains

n
* +n+1
fi = yn+1 − y0 = ∆−1 fi i=0
= yi ]n+1
i=0 where ∆yk = fk .
i=0
 205
One can verify that by adding the equations (6.6) from some point i = m to n,
one obtains the more general result
n
* +n+1
fi = yn+1 − ym = ∆−1 fi i=m
= yi ]n+1
i=m . (6.7)
i=m

Example 6-4.
Evaluate the sum

S = 1 · 2 + 2 · 3 + 3 · 4 + · · · + n(n + 1)

Solution: Let fk = k(k + 1) = k 2 + k and show one can write fk as the factorial
function fk = (k + 1)[2] . Therefore,
n
* *n ,n+1
[2] −1
+n+1 (i + 1)[3] (n + 2)[3] 2[3]
S= fi = (i + 1) = ∆ fi i=1 = = −
i=1 i=1
3 i=1 3 3

1
which simpliÞes to S = n(n + 1)(n + 2).
3

Difference Equations with Constant Coefficients

Difference equations arise in a variety of situations. The following are some

examples of where difference equations arise in applications. In assuming a
power series solution to differential equations, the coefficients must satisfy certain
recurrence formula which are nothing more than difference equations. In the
study of stability of numerical methods there occurs difference equations which
must be analyzed. In the computer simulation of various types of real-world
processes, difference equations frequently occur. Difference equations also are
studied in the areas of probability, statistics, economics, physics, and biology. We
begin our investigation of difference equations by studying those with constant
coefficients as these are the easiest to solve.

Example 6-5.
Given the difference equation

yn+1 − yn − 2yn−1 = 0

with the initial conditions y0 = 1, y1 = 0. Find values for y2 through y10 .

206
Solution: In the given difference equation, replace n by n + 1 in all terms, to
obtain
yn+2 = yn+1 + 2yn ,

then one can verify

n = 0, y2 = y1 + 2y0 = 2
n = 1, y3 = y2 + 2y1 = 2
n = 2, y4 = y3 + 2y2 = 6
n = 3, y5 = y4 + 2y3 = 10
n = 4, y6 = y5 + 2y4 = 22
n = 5, y7 = y6 + 2y5 = 42
n = 6, y8 = y7 + 2y6 = 86
n = 7, y9 = 78 + 2y7 = 170
n = 8, y10 = y9 + 2y8 = 342.

The study of difference equations with constant coefficients closely parallels

the development of ordinary differential equations. Our goal is to determine
functions yn = y(n), deÞned over a set of values of n, which reduce the given
difference equation to an identity. Such functions are called solutions of the dif-
ference equation. For example, the function yn = 3n is a solution of the difference
equation yn+1 − 3yn = 0 because 3n+1 − 3 · 3n = 0 for all n = 0, 1, 2, . . .. Recall that
for linear differential equations with constant coefficients we assumed a solution
y(x) = exp(ωx). We did this to obtain the characteristic equation and charac-
teristic roots associated with the differential equation. When x = n, we obtain
y(n) = yn = exp(ωn) = λn , where λ = exp(ω) is a constant. This suggests in our
study of difference equations with constant coefficients that we should assume a
solution of the form yn = λn , where λ is a constant. Analogous to ordinary linear
differential equations with constant coefficients, we Þnd that a linear, nth-order,
homogeneous difference equation with constant coefficients has associated with
it a characteristic equation with characteristic roots λ1 , λ2 , . . . , λn . The character-
istic equation is found by assuming a solution yn = λn , where λ is a constant.
The various cases that can arise are illustrated by the following examples.