Difference Equations

First-order linear difference equations

∆y = a(n)y + b(n) (∗)

A function y = f (n), Df ⊂ IN0 , is called a solution of the difference equation

(∗) if ∆f (n) = a(n)f (n) + b(n) ∀ n ∈ Df , where ∆y = y(n + 1) − y(n) =
f (n + 1) − f (n).
• If {a(n)} and {b(n)} are sequences of real numbers, then (∗) has the
solution


n−1

n−2 
n−1
y = f (n) = y0 · [a(k) + 1] + b(k) · [a(l) + 1] + b(n − 1)
k=0 k=0 l=k+1

Here f (0) = y0 ∈ IR can be chosen arbitrarily, while

/
n−1 
[a(0) + 1] · . . . · [a(n − 1) + 1] if n = 1, 2, . . .
[a(k) + 1] :=
1 if n = 0
k=0

/
n−1 
[a(k + 1) + 1] · . . . · [a(n − 1) + 1] if n = k + 2, . . .
[a(l) + 1] :=
1 if n = k + 1
l=k+1

In the special case a(n) ≡ a = const, b(n) ≡ b = const the solution of the
difference equation (∗) has the form

⎧
⎪
⎪ /
n−1
⎪
⎪ y · [a(k) + 1] if b(n) ≡ b = 0
⎪
⎪ 0
⎪
⎪
⎪
⎪
k=0
⎪
⎪ 
n−1
⎪
⎪
⎪
⎪ y (a + 1) n
+ b(k)(a + 1)n−1−k if a(n) ≡ a
⎪
⎨
0
k=0
y = f (n) = y0 (a + 1) n
if a(n) ≡ a, b(n) ≡ 0
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ (a + 1)n − 1
⎪
⎪ y (a + 1)n
+ b · if a(n) ≡ a = 0, b(n) ≡ b
⎪
⎪
0
a
⎪
⎪
⎪
⎩y +b· n
0 if a(n) ≡ 0, b(n) ≡ b
98 Difference equations

Economic models

y(n) – national income, n = 0, 1, 2, . . .

c(n) – consumption, n = 0, 1, 2, . . .
s(n) – sum of savings, n = 0, 1, 2, . . .
i(n) – investments, n = 0, 1, 2, . . .

Growth of national income according to Boulding

Model assumptions:
y(n) = c(n) + i(n), c(n) = α + βy(n), ∆y(n) = γi(n)

α – part of consumption independent of income, α ≥ 0

β – factor of proportionality for income dependent consumption, 0 < β < 1
γ – multiple of investments by which the national income changes, γ > 0

∆y(n) = γ(1 − β)y(n) − αγ, n = 0, 1, 2, . . . Boulding’s model

 
α α n
Solution: y = f (n) = + y0 − (1 + γ(1 − β))
1−β 1−β

• Under the assumption y(0) = y0 > c(0) the function y = f (n) is strictly
increasing.

Growth of national income according to Harrod

Model assumptions:
s(n) = αy(n), i(n) = β∆y(n), i(n) = s(n)

αy(n) – saved part of national income, 0 < α < 1

β – factor of proportionality between investments and increase of
national income, β > 0, β = α

Harrod’s model
α
∆y(n) = y(n), y(0) = y0 , n = 1, 2, . . .
β
 n
α
This model has the solution: y = f (n) = y0 ·
β
 Linear second-order difference equations 99

Ezekid’s cobweb model

Assumptions:
d(n) = α − βp(n), d(n) = n d(n) – demand,
q(n + 1) = γ + δp(n) p(n) – price
α > 0, β > 0, γ > 0, δ > 0 q(n) – supply

It is assumed that supply and demand are in equilibrium.

 
α−γ δ cobweb
∆p(n) = − 1+ p(n), p(0) = p0 , n = 1, 2, . . .
β β model
  n
α−γ α−γ δ
Solution: y = p(n) = + p0 − −
β+δ β+δ β

α−γ
• The quantity p(n) oscillates around the constant value p∗ = . For
β+δ
δ ≥ β the solution diverges, for δ < β the solution converges to the equilibrium
price p∗ .

c, d c, d c(n+1)
d(n) s
d(n)
c(n+1)
s sc(1)
d(1) s s
c(1) d(1)
s s sd(0)
s c(2) s s d(2)
c(2) s s

s d(0)

p(1) p∗ p(0) p(1) p(0) p(2)

Convergence Divergence

Linear second-order difference equations

An equation of the form

∆2 y + a∆y + by = c(n), a, b, c ∈ IR (∗)

is called a linear second-order difference equation with constant coefficients.

The term ∆2 f (n) := f (n+2)−2f (n+1)+f (n) is the second-order difference.
100 Difference equations

• If c(n) = 0 ∀ n = 0, 1, 2, . . ., then the equation is called homogeneous,

otherwise it is called inhomogeneous.
• A function f with Df ⊂ {0, 1, 2, . . .} is said to be a solution of the equation
(∗) if ∆2 f (n) + a∆f (n) + bf (n) = c(n) ∀ n ∈ Df .
• The general solution of the linear inhomogeneous difference equation (∗)
is the sum of the general solution of the associated homogeneous difference
equation ∆2 y + a∆y + by = 0 and any special solution of (∗).

General solution of the second-order homogeneous difference equa-

tion
Consider the characteristic equation λ2 + aλ + b = 0.

a 1 2
Its solution is determined from the formula λ1,2 = − ± a − 4b. De-
2 2
pending on the discrimant D = a − 4b it can have two real, one real double
2

or two conjugate complex solutions. To represent the general solution of the

homogeneous difference equation associated with (∗) one has to distinguish
between three cases, where C1 , C2 are arbitrary real constants.
1 √  1 √ 
Case 1 D > 0 : λ1 = −a + D , λ2 = −a − D
2 2

Solution: y = f (n) = C1 (1 + λ1 )n + C2 (1 + λ2 )n

a
Case 2 D = 0 : λ1 = λ2 =: λ = − (a = 2)
2

Solution: y = f (n) = C1 (1 + λ)n + C2 n(1 + λ)n

a 1√
Case 3 D < 0 : α := − , β := −D
2 2
Solution:
n n
y = f (n) = C1 [(1 + α)2 + β 2 ] 2 cos ϕn + C2 [(1 + α)2 + β 2 ] 2 sin ϕn
β π
where tan ϕ = (α = −1) and ϕ = (α = −1).
1+α 2

General solution of the second-order inhomogeneous difference

equation
The general solution of the inhomogeneous equation is the sum of the general
solution of the homogeneous equation and a special solution of the inhomo-
geneous equation (∗). The latter can be obtained e. g. by means of the ansatz
method, where the corresponding ansatz functions depend on the concrete
 Economic models 101

structure of the right-hand side c(n). The unknown coefficients involved are
determined by comparison of coefficients.

right-hand side trial solution

c(n) = ak nk + . . . + a1 n + a0 C(n) = Ak nk + . . . + A1 n + A0

c(n) = a cos ωn + b sin ωn C(n) = A cos ωn + B sin ωn

(α = 0 or β = ω; see Case 3 on p. 100)

Economic models

y(n) – national income c(n) – consumption

i(n) – private investments H – public expenditure

Model assumptions (n = 0, 1, 2, . . .)
y(n)=c(n) + i(n) + H the national income splits up into con-
sumption, private investments and public
expenditure
c(n)=α1 y(n − 1) 0 < α1 < 1; the consumption is proportional
(multiplicator α1 ) to the national income of
the previous period
i(n)=α2 [c(n) − c(n − 1)] α2 > 0; the private investments are propor-
tional (accelerator α2 ) to the increase of the
consumption

Samuelson’s multiplicator-accelerator model

∆2 y + (2 − α1 − α1 α2 )∆y + (1 − α1 )y = H

Solution for H n
y = f (n) = + (α1 α2 ) 2 (C1 cos ϕn + C2 sin ϕn)
α1 ≤ α2 < 1: 1 − α1

y
• The solution f oscillates
with decreasing amplitude rr
r r
H r f (n)
around the limit .
1 − α1 r
r
r rrrrrrrr
rr
H
r
rr
1−α1
rr
rrr

n
102 Difference equations

Linear difference equations of n-th order with constant coefficients

yk+n + an−1 yk+n−1 + . . . + a1 yk+1 + a0 yk = c(k) (k ∈ IN) (1)

• A linear difference equation of the form (1) with constant coefficients
ai ∈ IR, i = 0, 1, . . . , n − 1, is of n-th order if a0 = 0.
• The difference equation of n-th order (1) has exactly one solution y k = f (k)
if the initial values for n successive values k are given.
• If f1 (k), f2 (k),. . . , fn (k) are arbitrary solutions of the homogeneous linear
difference equation
yk+n + an−1 yk+n−1 + . . . + a1 yk+1 + a0 yk = 0, (2)
then the linear combination
f (k) = γ1 f1 (k) + γ2 f2 (k) + . . . + γn fn (k) (3)
with (arbitrary) constants γi ∈ IR, i = 1, . . . , n, is also a solution of the
homogeneous difference equation (2).
• If the n  solutions f1 (k), f2 (k),. . . , fn (k) of (2) form a fundamental sys-
 f1 (0) f2 (0) . . . fn (0) 
 

tem, i. e.  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  = 0, then (3) is the general
 f1 (n−1) f2 (n−1) . . . fn (n−1) 
solution of the homogeneous difference equation (2).
• If yk,s is a special solution of the inhomogeneous linear difference equation
(1) and yk,h is the general solution of the associated homogeneous linear
difference equation (2), then for the general solution of the inhomogeneous
linear difference equation (1) the representation yk = yk,h + yk,s holds.

General solution of the n-th order homogeneous difference equation

Solve the characteristic equation λn + an−1 λn−1 + . . . + a1 λ + a0 = 0 .
Let its solutions be λ1 , . . . , λn . Then the fundamental system consists of n
linearly independent solutions f1 (k), . . . , fn (k), whose structure depends on
the kind of the solutions of the characteristic equation (analogously to

second-order difference equations, p. 100).

Special solution of the n-th order inhomogeneous difference equa-

tion
To find a special solution of the inhomogeneous difference equation (1), in
many cases the ansatz method is successful, where the ansatz function is cho-
sen in such a way that it corresponds to the right-hand side with respect to
structure (
second-order difference equation, p. 100). The unknown coeffi-
cients contained are determined by substituting the ansatz function into (1)
and making a comparison of coefficients.