Linear Difference Equations

Posted for Math 635, Spring 2012.

Consider the following second-order linear difference equation

f (n) = af (n 1) + bf (n + 1), K < n < N, (1)

where f (n) is a function defined on the integers K n N , the value N can be chosen to be
infinity, and a and b are nonzero real numbers. Note that if f satisfies (1) and if the values f (K),
f (K + 1) are known then f (n) can be determined for all K n N recursively via the formula
1 
f (n + 1) = f (n) af (n 1) .
b
Note also that if f1 and f2 are two solutions of (1), then c1 f1 + c2 f2 is a solution for any real
numbers c1 , c2 . Therefore, the solution space of (1) is a two-dimensional vector space and one basis
for the space is {f1 , f2 } with f1 (K) = 1, f1 (K + 1) = 0 and f2 (K) = 0, f2 (K + 1) = 1.
We will solve (1) by looking for solutions of the form f (n) = n , for some 6= 0. Plugging n
into equation (1) yields
n = an1 + bn+1 , K < n < N,
or
= a + b2 b2 + a = 0.
Solving this quadratic gives
1
1 4ba
= . (2)
2b
There are two cases that need handling based upon whether or not the discriminant, 1 4ba,
is zero.
Case 1: If 1 4ba 6= 0, we find two solutions, 1 and 2 , and see that the general solution to the
difference equation (1) is
c1 1n + c2 2n ,
with c1 , c2 found depending upon the boundary conditions. If 1 4ba < 0, then the roots are
complex and the general solution is found by switching to polar coordinates. That is, we let
= rei , and find
f (n) = rn ein = rn cos(n) irn sin(n),
are solutions, implying both the real and imaginary parts are solutions. Therefore, the general
solution is
c1 rn cos(n) + c2 rn sin(n),
with c1 , c2 found depending upon the boundary conditions.
Case 2: If 1 4ba = 0, we only find the one solution, f1 (n) = (1/2b)n by solving the quadratic.

1
However, let f2 (n) = n(2b)n . We have that

af2 (n 1) + bf2 (n + 1) = a(n 1)(2b)(n1) + b(n + 1)(2b)(n+1)

 n  
1 1
= a(n 1)2b + b(n + 1)
2b 2b
 n  
1 1 1
= (n 1) + (n + 1) (remember, 4ab = 1)
2b 2 2
 n
1
= n
2b
= f2 (n).

Note that f2 is obviously linearly independent from f1 . Thus, when 4ab = 1, the general form of
the solution is  n  n
1 1
f (n) = c1 + c2 n .
2b 2b
with c1 , c2 found depending upon the boundary conditions.
Example 1. Find a function f (n) satisfying
1
f (n) = 2f (n 1) + f (n + 1), 0 < n < ,
10
with f (0) = 8, f (1) = 2.
Solution. Here, a = 2 and b = 1/10. Therefore, plugging into (2) gives

= 5 5,

and the general solution is

 n  n
f (n) = c1 5 + 5 + c2 5 5 .

Using the boundary conditions yields

8 = f (0) = c1 + c2

2 = f (1) = c1 (5 + 5) + c2 (5 5),

which has solution c1 = 4 19 5/5, c2 = 4 + 19 5/5. Thus, the solution to the problem is
! !
19 5  n 19 5  n
f (n) = 4 5+ 5 + 4+ 5 5 .
5 5


Some of the most important difference equations we will see in this course are those of the form

f (n) = pf (n 1) + qf (n + 1), with p + q = 1, p, q 0.

These will arise when studying random walks with p and q interpreted as the associated probabilities
of moving right or left. Supposing that p 6= q, the roots of the quadratic formula (2) can be found:
p p
1 (1 2p)2
 
1 1 4(1 p)p 1 |q p| p
= = = 1, .
2q 2q 2q q

2
Thus, the general solution when p 6= 1/2 is
 n
p
f (n) = c1 + c2 .
q

For the case that p = q = 1/2, the only root is 1, hence the general solution is

f (n) = c1 + c2 n.

We analyzed only second-order linear difference equations above. However, and similar to the
study of differential equations, higher order difference equations can be studied in the same manner.
Consider the general kth order, homogeneous linear difference equation:

f (n + k) = a0 f (n) + a1 f (n + 1) + + ak1 f (n + k 1), (3)

where we are given f (0), f (1), . . . , f (k 1). Then, again, we may solve for the general f (n) re-
cursively using (3). We look for solutions of the form f (n) = n , which is a solution if and only
if
k = a0 + a1 + + ak1 k1 .
If there are k distinct roots of the above equation, then we automatically get k linearly independent
solutions to (3). However, if a given root is a root with a multiplicity of j, then

n , nn , . . . , nj1 n ,

are linearly independent solutions. We can then use the given initial conditions to find the desired
particular solution.