Linear Difference

Equations

BLOCK 6
DIFFERENCE EQUATIONS

273
Difference Equations
BLOCK 6 INTRODUCTION
The final Block of the course, titled Difference Equations, slightly changes
the approach from what had been followed in Blocks 3, 4 and 5, in that the
present Block does not involve a discussion of functions and variables that are
continuous. Rather the block discusses variables and functions that are discrete.
Specifically, it discusses equations in which values of variables are related to
lagged values of themselves. Actually you have to understand that basically we
are looking for a way to depict, model and study the evolution of economic
variables over time. We look at a variable say x, and see that if its value at time
t (time measured as discrete periods)depends on past values. Difference
equations study just this.

Te Block has two units, Unit 15, titled Linear Difference Equations, and Unit
16, titled Non-Difference Equations. It should be clear from the titles that unit
15 discusses difference equations that are linear. The unit discusses the concept
of finite differences. The unit explains basic operators like lag and backward-
shift operators. You learn about various types of equations like homogeneous
and non-homogeneous difference equations; autonomous and non-autonomous
equations. The unit studies linear difference equations of first- and second-
degree, with relevant economic applications.

The final unit of the block, also the final unit of the course, Unit 16, Non-
Difference Equations, discusses about difference equations that are non-
linear. The unit discusses phase diagram and qualitative analysis. Solution
methods for non-linear difference equations are discussed. Non-linear
difference equations are applied to certain areas in economics like cycles,
chaos, and so on

274
 Linear Difference
UNIT 15 LINEAR DIFFERENCE Equations

EQUATIONS*
Structure
15.0 Objectives
15.1 Introduction
15.2 Preliminary Concepts
15.3 First Order Difference Equations
15.3.1 The Iterative Method
15.3.2 The General Method
15.4 Second Order Difference Equations
15.5 Economic Applications
15.5.1 The Cobweb Model
15.5.2 Samuelson Multiplier-Accelerator Interaction Model
15.6 Let Us Sum Up
15.7 Answers/Hints to Check Your Progress

15.0 OBJECTIVES
After studying the unit, you should be able to:
• understand finite differences;
• explain discrete economic processes;
• define a linear difference equation;
• identify the order of a difference equation;
• explain linear difference equations of first and second degree;
• differentiate between a homogeneous and non-homogeneous difference
equation;
• describe methods to solve first order and second order difference
equations; and
• apply difference equations to depict, set up and solve some dynamic
economic models in discrete time.

15.1 INTRODUCTION
By now you are familiar with differential calculus and integral calculus and
have learned how to solve problems in economics using techniques of
differential calculus. You have also seen how integration is the reverse of
differentiation. If we want to find whether a given function is increasing or
decreasing, sign of the differential coefficient of the dependent variable with
respect to the independent variable helps in determining it. If we want to decide
about the maximum or minimum values of a function, again we take the help of
the differential co-efficients of various orders and so on. So, the idea of

* Contributed by Shri Jagmohan Rai 275

Difference Equations differential coefficient is central to all the techniques developed in differential
calculus. But if y is a function of x, we can find its derivative only if x is a
continuous variable, that is, only if the change in x is infinitesimally small. In
such a case, y is defined for every value of x in an interval. But this may not to
the case always.

In Economics, we do come across situations in which the independent variable

is not continuous but discrete is nature. That is, x may take two values, say, 2
and 4, without taking all the values between 2 and 4. In such a case, if y is a
function of x, then y is not defined in the entire interval (2, 4), but only for the
values x = 2 and x = 4. For example, suppose we are studying the savings of a
family and collect the data at fixed time intervals, say every year. So if S =
S(t), where S is the savings at the end of time interval t, we get S as a function
of time t which takes values 0, 1, 2, ....,etc. and hence is a discrete variable.
We may be interested in studying changes in the dependent variable due to
changes in the independent variable, in such cases as well. The concept of
derivative no longer works here, as the independent variable involved takes
discrete and not continuous values. The study of relations that exist between
the values assumed by a function whenever the independent variable changes
by finite jumps is the subject matter of the calculus of finite differences. In
this unit and the next, we introduce difference equations and give methods to
solve them. We also study some situations in economics where difference
equations may prove to be useful. The present unit deals with difference
equations.

15.2 PRELIMINARY CONCEPTS

We will be considering a function, with a discrete and not continuous
independent variable, that is, in a function where y will be the dependent and x
the independent variable, x will be taking integer values. The pattern of change
in y, the dependent variable will then be determined by “differences” and not
derivatives or differentials of the function y(x).

Suppose that y = f (x), that is, y is a function of x, where x takes equidistant

integer values x0, x0+h, x0+2h,…, x0+nh.

Then f (x0), f (x0+h), f (x0+2h),…, f (x0+nh) will be the corresponding y values

denoted by y0, y1,y2,..., yn.

The differences, y1− y0, y2 − y1,…,yn− yn–1 are known as the first forward
differences of the function y = f (x), denoted by ∆ y0, ∆y1,…,∆yn respectively.
These differences are what we were looking for to represent the change in y
with respect to the discrete variable x.

Generally, first forward difference is defined as:

∆yn = yn+1−yn, or
∆f (x) = ∆f (x + h) −f (x), where h is the interval of differencing.
∆y or ∆f (x) is basically the difference between two consecutive values of the
discrete variable x.
For example, consider a sequence: −5, −1, 3, 7,….., (4x + 3),….

276
Here, we have y or f (x) = 4x + 3 and h = 4 (calculated by subtracting any two Linear Difference
consecutive values of the sequence). Then, the first forward difference will be Equations
given by:
∆y or ∆f (x) = ∆f (x + h) −f (x)
= [4(x +h) + 3] – [4x + 3]
= 4x + 4h + 3– 4x – 3
= 4h = 16 [∵ h = 4 (calculated)]
We define second forward difference,∆2ynas ∆(∆yn) so that
∆ 2 yn = ∆(∆yn ) =∆ (y n+1 − yn )
= ( yn + 2 − yn +1 ) − ( yn +1 − yn )
= yn + 2 − 2 yn +1 + yn

Similarly, we can define forward difference of higher order. Since our basic
purpose is to study problems in Economics and since we are interested in
finding a time path of a variable y from some given pattern of change over
time, we take t as independent variable and write ∆yt = yt +1 − y.t . Thus, in the
proceeding example, we may write ∆yt = 16. Such an equation is called a
difference equation. The order of a difference equation is the order of the
highest difference it contains. For example, ∆2yt= 4 is a difference equation of
order 2.
The difference equation, ∆yt = 16 …(a)
can also be rewritten asyt+1−yt = 16 or yt+1 = yt + 16 …(b)
Similarly, the difference equation ∆yt = −0.3 yt …(c)

can also be rewritten asyt+1−yt = −0.3 yt or yt+1 = 0.7 yt …(d)

A difference equation is homogeneous if its constant term, the term that does
not contains y is zero, otherwise it is non-homogenous. Thus, we can define a
difference equation as an equation involving an independent variable, a
dependent variable and the successive difference of the dependent variable.
Also, we have seen that the successive difference of a dependent variable can
be expressed in terms of the successive values of the dependent variable. For
example equation (a) can be rewritten as equation (b) and equation (c) as
equation (d). So, we may also define a difference equation as an equation
which expresses a relation between an independent variable and the successive
values of the dependent variable. Difference equations expressed in this form
[i.e. in the form (b), (d)] are more convenient to deal with.Note that ∆ 2 yt = 4
can be rewritten as yt + 2 − 2 yt +1 + yt = 4 .

A solution of a difference equation is a relation between the independent

variable and the dependent variable satisfying the given equation. The general
solution of a difference equation of order n contains n arbitrary constants. For
example ∆2yt= 0 is a homogeneous difference equation of order 2 and ∆yt = 4 is
a non-homogeneous difference equation of order 1. A difference equation is
said to be linear if every y present in the equation is at most of degree 1. We
will first learn how to solve first order difference equations. Let us discuss
different types of difference equations and their solution methods one by one.
277
Difference Equations
15.3 FIRST ORDER DIFFERENCE EQUATIONS
In this sub-section we will learn to solve difference equations of order 1. We
will first discuss the iterative method for solving a first order difference
equation.
15.3.1 The Iterative Method
This method is explained with the help of an example.
Consider the difference equation ∆yt = 4
As seen earlier it can be rewritten as
yt+ 1 = yt + 4 …(i)
Putting t = 0, 1, 2, 3,.....in equation (i) we get,
y1 = y0 + 4
y2 = y1 + 4 = (y0 + 4) + 4 = y0 + 2(4)
y3 = y2 + 4 = y0 + 2(4) + 4 = y0 + 3(4)
and so on.
In general, we have
yt = y0+ 4t …(ii)
Now suppose y0 = 20. Inserting y0 value in equation (ii), we get
yt = 20 + 4t …(iii)

as the solution of the given difference equation ∆yt = 4, subject to the assumed
value of y0. Note that the solution in equation (iii) gives us the value of y
corresponding to any value of t. If the initial value (y0) is not specified, the
solution would be obtained in terms of y0.

.For example, the difference equation ∆yt= − 0.3 yt …(iv)

can be written as yt+1 – yt = − 0.3 yt or yt+1 = 0.7yt …(v)
and can be solved using the iterative method as follows:
Putting t = 0, 1, 2, ....etc., in equation (v), we get
y1 = 0.7 y0
y2 = 0.7 y1 = (0.7) 2 y0
y3 = (0.7)3 y0

and so on. In general, we get yt = (0.7)t y0

as the solution to the difference equation (iv) in terms of y0 .

We will now discuss the general method for solving a first order difference
equation.

278
15.3.2 The General Method Linear Difference
Equations
Consider the first order difference equation
yt+1 + a0y1 = a …(vi)

wherea0 and a are constants. The general solution of the difference equation
(vi) consists of two parts, a particular integral yp and a complementary function
yc. The complementary function yc is the general solution of the homogeneous
difference equation (vii) corresponding to equation (vi).
yt+1 + a0yt = 0 …(vii)
Whereas, a particular integral yp is any solution of the given non-homogenous
difference equation (vi). The general solution of (vi) is the sum of ycandyp.
This method is called the general method because it helps us in solving the
most general linear difference equation of order 1. Not only that, the method
applies to linear difference equations of higher order as well. Since we have
also to study second order linear difference equations and their solutions, and
the method is similar for the first order and second order difference equations
(in fact for linear difference equations of any order for that matter), we
postpone its discussion for a while to avoid unnecessary repetition.
Check Your Progress 1
1) What is the difference between homogeneous and non-homogeneous
difference equations?
……………………………………………………………………………..
……………………………………………………………………………..
……………………………………………………………………………..
……………………………………………………………………………..
……………………………………………………………………………..
2) Solve the following linear difference equation by Iterative method:
i) yt +1 − 6 yt = 0

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
ii) ∆ =0
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
iii) = 10 +
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
279
Difference Equations iv) ∆ =2
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………

15.4 SECOND ORDER DIFFERENCE EQUATIONS

A difference equation not involving successive values of yt greater than yt+2is
said to be of order 2. Some examples of second-order difference equations are:
i) ut+ 2 − 3ut + 1 + 5ut= 10t
ii) yt + 2 − 4 yt = 0

yt + 2 − 3 yt +1 + 4 yt = 6 + t
iii)

iv) yt + 2 + yt +1 − 2 yt = 12t

In general, any second order linear, non-homogenous difference equation with

constant coefficient takes the form,
yt + 2 + a1 yt +1 + a2 yt = φ (t ) …(viii)

wherea1 and a2 are constants and φ (t ) is a function of t only. If φ (t ) = 0, we

get
yt + 2 + a1 yt +1 + a2 yt = 0 … (ix)

which is a second order linear, homogeneous difference equation with constant

coefficients.
A general solution of (viii) is, again, given by yc + yp, where yc is the general
solution of (ix) and y p is any solution of (viii). We will first learn how to
solve (ix).This method is quite general in the sense that it can be used to solve
any homogeneous linear difference equation of any order including equation
(vii) of order 1. The method is as follows:
Step 1: In the difference equation (ix) replace yt by 1, yt+1 by a variable, say,
b, and yt+2 byb2 so as to obtain the following characteristic equation:
b2 +a1b +a2 = 0 …(x)
Step 2: Solve the characteristic equation (x). The values of b so obtained are
called characteristic roots.
Step 3: The general solution of (ix) can be written using the characteristic
roots obtained in step 2. The solution depends on the nature of the
characteristic roots, giving rise to three possibilities.
Case I: When the characteristic roots obtained in step 2 are real and
distinct, say b1 and b2, then the general solution to (ix)
will be written as:
t
y = Ab
1 1
+ A2b2t …(xi)
280
 Case II: When the characteristic roots obtained in step 2 are real but Linear Difference
repeated say b1 and b2 = b1, then the general solution to (ix) Equations
will be written as:
y = ( A1 + t A 2 )b1t …(xii)

Case III: When the characteristic roots obtained in step 2 are complex,
say b1 = α + i β , b2 = α − i β , then the general solution to
(ix) will be written as:
y = A1(α + iβ)t + A2(α−
−iβ)t … (xiii)
This can also be written as

y = rt(C1cosθt+ C2sinθt) …(xiv)

Where, r= + , θ = tan−1 , C1 = A1 + A2, C2 = i(A1 −A2)

Note:
1. A1,A2, C1, C2 used above are all constants.
2) The roots in case III have been chosen to be α + iβ and α− −iβ. Here, α
andβ are real numbers. In fact, whenever a quadratic equation with real
coefficients has complex roots, they have to be of the form α ± iβ.
3) The case of complex repeated roots has not been discussed since we are
dealing with equations of order up to 2 only, and for an equation to
have repeated complex roots, its order must be at least 4.
4) When we are dealing with a non-homogeneous difference equation of
type (viii), then the general solution of the corresponding homogeneous
equation (ix) is called the complementary function of (viii) and is
denoted by yc. So in that case, let the solutions given in (xi), (xii), (xiii)
or (iv) be denoted by yc.
Let us take some examples to illustrate the procedure.
Example 1 Solve yt+2 – 4yt = 0.
Solution: The characteristic equation (with yt = 1 and yt+2 = b2) of the
difference equation is
b2 – 4 = 0 ⇒ b2 = 4 ⇒ b = ±2
Thus, solutions are b1 = 2 and b2 = – 2. These are real and distinct.
Therefore, the general solution to the given equation is:
y = A1 2t + A2 (–2)t, (where A1 and A2 are constants)
[using equation (xi)]
Example 2 Solve yt+2 + 10 yt+1 + 25yt = 0
Solution The characteristic equation (with yt = 1, yt+1 = b and yt+2 = b2) for this
difference equation is:
b 2 + 10b + 25 = 0
Solving this equation we get –5 as a repeated root. Therefore, the
general solution to the given equation is:
281
Difference Equations −5)t
y = (A1 + t A2) (− [using equation (xii)]

Example 3 Solve yt + 2 − 4 yt +1 + 13 yt = 0

Solution The corresponding characteristic equation of the given equation is

b 2 − 4b + 13 = 0
On solving, you will realize that its roots are complex given by,
b1 = 2 + 3i, b2 = 2 − 3i.

∴α = 2, β = 3. This implies that = √2 + 3 = √13

∴ The general solution to the given equation is:
3
y = √13 (C1cosθt+ C2 sin θt), where θ = tan−1
[using equation (xiv)]
Now we come to the general solutions of a non-homogeneous linear difference
equation of order 2, of the form (viii). As discussed earlier, the general
solution to equation (viii) will consist of two parts, yc and yp. Here, yc , the
complementary function, is the general solution of the corresponding
homogeneous difference equation (ix), and we have already learned how to
find it. Now we will learn to find y p , the particular integral for the difference
equation (viii). In fact, y p depends on the function φ (t ) on the right hand side
of the equation (viii). We will discuss methods for obtaining the particular
integral for some special types of φ (t) only. To be specific, we will be
discussing only the following cases:
Case 1: When φ (t ) is a constant function.

Case 2: When φ (t ) is any polynomial function.[Note that case 1 is just a

special case of case 2]
Case 3: When φ (t ) is an exponential function of the form at , for some
constant a.
Case 4: When φ (t) is a polynomial function multiplied by at for some constant
a.
We will discuss these cases one by one. Since we can choose any solution of
p
equation (viii) as y , we suggest methods for choosing the function which can
be tried for the particular integral in each of the above mentioned cases.
Case 1 When φ (t ) is a constant, say a, we choose

yt = k

as a possible solution of equation (viii), where k is a constant to be determined

by substituting yt = k in (viii).

Let us illustrate the procedure with the help of an example.

Example 4 Solve yt + 2 − 4 yt = 5

282
Solution: Note that we are being given a non-homogenous equation, and the Linear Difference
homogenous equation corresponding to this equation is Equations

yt + 2 − 4 yt = 0

which is the equation that has been solved in example 1. We already

know the characteristic roots of the homogenous equation, giving us
the complementary function as the general solution.

∴ = 2 + (−2)
Now we proceed to find the particular integral. Since φ (t ) = 5, in
this case we choose
yt = k

as a trial solution for finding the particular solution of the given

equation.
yt = k implies yt + 2 = k [y being a constant function]

Now substituting values of yt and yt+2 in our non-homogeneous

equation, we get ! − 4! = 5
$%
!=
&
$%
∴ ' = &

is a particular integral y p of the given equation. Thus, the

complete solution of equation yt + 2 − 4 yt = 5 is given by

yt =yc + yp
%
= A1 2t + A2 (–2)t−
&

Case 2: When φ (t ) is a polynomial of degree n in t, we choose the general

polynomial yt = p0 + p1t +……pntn
of degree n as a possible candidate for the particular integral and find the
p p ,....., pn
constants 0, 1 by substituting in the given difference equation.
Example 5 Solve yt + 2 + 10 yt +1 + 25 yt = 1 + t 2

Solution The complementary function yc = (A1 + t A2) (−−5)t, has already been
obtained in Example 2. To get the particular integral, we choose the
general polynomial
yt = p0 + p1t +p2t2
since we have φ (t ) = 1 + t 2 , a polynomial of degree 2. The
difference equation in the our example contains, besides yt, the
terms yt +1 and yt + 2. . We thus find these values from our general
polynomial:
yt+1 = p0 + p1(t + 1) + p2(t + 1)2
283
Difference Equations = (p0 + p1+ p2) + (p1 + 2 p2)t + p2 t2
yt+2 = p0 + p1(t + 2) + p2(t + 2)2
= (p0 +2 p1+ 4p2) + (p1 + 4 p2)t + p2 t2

Now, on substituting the values of yt, yt+1 and yt+2 in our difference
equation, we get
[(p0 +2 p1+ 4p2) + (p1 + 4 p2)t + p2 t2]
+ 10 [(p0 + p1+ p2) + (p1 + 2 p2)t + p2 t2] + 25 [p0 + p1t +p2t2] = 1 + t2
Collecting the coefficients of like terms on L.H.S, we get
(36p0 + 12 p1 + 14 p2) + (36 p1 + 24 p2)t + (36 p2)t2 = 1 + t2
Now, equating the coefficients of both the sides,
36 p0 + 12 p1 + 14 p2 = 1
36 p1 + 24 p2 = 0
36 p2 = 1
$ $ )
Solving these equations, we get p2 = &(, p1 = &) , p0 = *+ +
$ ) $
yp = + t + t2
*+ + &) &(
Thus we get

as a particular integral for the given difference equation. Thus, a

complete solution to the given difference equation
yt + 2 + 10 yt +1 + 25 yt = 1 + t 2 is

yt = yc + yp
$ ) $
−5)t+ [ + t + t2]
*+ + &) &(
yt= (A1 + t A2) (−

Case 3: When φ (t ) = a t for some constant a.

In this case the choice for the trial functions depends upon whether a is a root
of the characteristic equation of the given difference equation or not. If a is not
a root of the characteristic equation, then yt = ca t , where c is a constant to be
determined, works as our trial function.
Whereas, if a is a root of the characteristic equation, then we check if it is
repeating or not. If it is repeating m times, we choose yt = ct m at as our trial
equation. If a is a root of the characteristic equation and is not repeated, then m
= 1, and we take yt = cta t as our trial equation. If it is a root repeated twice,
then m = 2, and we take yt= ct2at as our trial equation, and so on.
We take two examples to illustrate.
Example 6 Solve yt + 2 − 5 yt +1 + 6 yt = 7t

Solution It is easy to see that in this case, the characteristics equation to the
given difference equation will have b1 = 3 and b2 = 2 as the real and
distinct characteristic roots. Thus the complimentary function (C.F)
will be given by
284
 yc=A13t + A22t [using equation (xi)] Linear Difference
Equations
To find y p , note that φ (t ) = 7t and 7 is not a root of the
characteristic equation, therefore we choose the following as our
trial function:
yt = c7t

Now, we find yt+1 and yt+2 from the trial function.

yt +1 = c.7t +1

yt + 2 = c.7t + 2

Now, on substituting the values of yt,yt+1 and yt+2 in our difference

equation, we get
c.7t+2– 5.c.7t+1 + 6. c.7t = 7t

Dividing both sides by 7t , we get

49 c – 35 c + 6 c = 1 ⇒
,
c=

∴yp = ,
. 7tis the required particular integral (PI).

Now, the complete solution of the difference equation

yt + 2 − 5 yt +1 + 6 yt = 7t turns out to be:

yt = CF + PI
1 t
=A1 3t + A2 2t + 7
20
Example 7 Solve yt+2 – 7 yt+1 + 10yt= 3×5t
Solutions In this case, the characteristic equation has solutions 2 and 5.
Therefore, the complimentary function (CF) becomes:
yc=A12t + A25t
Now ∅(/) =3×5t, with 5 occurring in 5t being one of the roots of the
characteristic equation, but non-repeating. Therefore, we take
yt = cta t as our trial equation, with m = 1. Now,

yt=c.t.5t
⇒ yt+1 = c.(t+1).5t+1
⇒ yt+2 = c.(t+2).5t+2
Now, on substituting the values ofyt,yt+1 and yt+2 in our difference
equation, we get
c.(t+2).5t+2– 7.c.(t+1).5t+1+ 10. c.t.5t = 3×5t
Dividing both sides by 5t , we get
25.c.(t+2)– 35.c.(t+1) + 10. c.t = 3
⇒ (25c – 35c + 10c).t + 50c – 35c = 3
285
 ⇒ 15c = 3 ⇒ c =
Difference Equations
%

∴yp = t 5tis the required particular integral (PI).

%

Now, the complete solution of the difference equation yt+2 – 7 yt+1 +

10yt= 3×5t turns out to be:
yt = C.F . + P.I
1
=A1 2t + A2 5t + t 5t
5
Case 4: When ∅(/) = (a0+ a1t +…+ antn) at, where a, a0,…an are constants. In
this case ∅(/) is a product of functions considered in case 2 and case 3. The
trial solution to be considered for y p is also a product of the solutions
considered in case 2 and case 3, respectively.
Thus if ∅(/) = (a0+ a1t +…+ antn) at, we choose yt = ( p° + p1t + .... pnt n )a t as our
trial equation, if a is not a root of the characteristic equation of the associated
m n t
linear homogeneous difference equation and yt = t ( p° + p1t + .... pnt )a if a is
a root of the characteristic equation repeated m times. For example, if the
difference equation is
yt + 2 − 5 yt +1 + 6 yt = 3t 2 7t

then we shall choose, yt= (p0 + p1t + p2 t2) 7t as the trial function for the
particular integral because ∅(/) in this case is 3t 2 7 t , which is the product of a
second degree polynomial and 7t , with 7 is not a root of the characteristic
equation of the associated homogeneous linear difference equation.
But if we have the difference equation yt + 2 − 14 yt +1 + 49 yt = 3t 2 7t

we shall choose yt = t 2 ( p0 + p1t + p2t 2 )7t

as the trial function, because in this case, the corresponding characteristic

equation b 2 − 14b + 49 = 0 has 7 as a root repeated twice. This necessitated the
induction of the factor t 2 in the trial equation. The values of the constantsp0,
p1,…,pnin the trial functions above can be obtained by substituting the value of
yt of the trial function, in the respective difference equations. The calculation
of the constants, of course, becomes more and more complicated with higher
values of m and n in our trial equations. Even the difference equations of the
type mentioned above need a lot of calculation. But the learner shall seldom
encounter problem with higher values of n and m. We advise the learner to
solve the above level of difference equations, along with some simple
problems of this type given in the exercise that follows this unit. The
techniques for solving difference equations of order 2, discussed here, are
general in nature and can be applied to linear difference equations of any order.
In particular, linear difference equations of order 1 can be easily solved using
these techniques. We illustrate our point with the help of some examples.
Example 8 Solve the difference equation yt +1 − 4 yt = 3

Solution In this case, the corresponding homogeneous difference equation is

286 yt +1 − 4 yt = 0
 and therefore the characteristic equation b − 4 = 0 has b = 4 as the Linear Difference
only root. Equations

∴ yc = A(4)t as the required complimentary function (C.F).

To find particular integral (P.I), we take yt = k as the trial function,

where k is to be determined, because ∅(/) = 3 is of the type
discussed in case 1.
Now, yt = k ⇒ yt+1 = k [yt being a constant function]
Substituting these values in the difference equation yt +1 − 4 yt = 3 ,
we get
k ‒ 4k = 3 ⇒k = ‒1
∴yp= ‒1 is the required particular integral (P.I).
Thus, the complete solution of the difference equation yt +1 − 4 yt = 3
will be given by
yt = C.F + P.I ⇒ yt = A(4)t ‒1
Example 9 Solve yt +1 − 2 yt = 12t

Solution Here ∅(/) = 12/ , a polynomial of degree 1.

∴we will take yt = p0 + p1t as our trial function.
Now, yt+1 = p0 + p1(t + 1)
On substituting the values of yt and yt+1 in the given difference
equation, we get
[p0 + p1(t + 1) ‒ 2(p0 + p1t)] = 12 t
⇒ (‒p0 + p1) ‒ p1t = 12t
Now, compare coefficients of both the sides:
‒ p1 = 12 ⇒ p1 = ‒12
⇒ p0 = p1 = ‒12
∴we get yp = ‒12 ‒12t as the particular integral (P.I)
Clearly, C.F. = A.2t
∴ The complete solution of the given equation is

yt = C.F . + P.I .
= A.2t − 12(1 + t )

Example 10 Solve yt +1 + 5 yt = 6t

Solution: In this case the characteristic equation of the associated

homogeneous difference equation has − 5 as the only root and
∅(/) = 6t is a function of the type discussed in case 3. Therefore,
we choose yt = c.6t as our trial equation because 6 is not a root of

287
Difference Equations the characteristic equation. This gives yt +1 = c.6t +1 . On substituting
the values of yt and yt+1is the given difference equation we get,
c.6t+1+ 5. c.6t = 6t
Dividing both sides by 6t, we get
6c + 5c = 1
1
⇒ c=
11
1 t
∴ The particular integral (P.I), y p = 6
11

Also, C.F ., yc = A(−5)t

t 1 t
∴ The complete solution is yt = C .F . + P.I = A( −5) + 6
11

Example 11 Solve yt +1 − 3 yt = 4.(3)t

Solution In this case complimentary function (C.F) = A.(3)t as 3 is the only

root of the characteristic equation and given that φ (t ) = 4.3t , we
choose yt = c.t.3t as our trial equation, because 3 occurring in 3t is a
non-repeating root of the characteristic equation.
yt = c.t.3t ⇒ yt+1 = c.(t + 1)3t+1

Now, on substituting the values of yt and yt+1 in the given difference

equation yt +1 − 3 yt = 4.(3)t , we get

c.(t + 1)3t+1‒ 3(c.t.3t) = 4(3)t

Dividing both the sides by 3t, we get
3c (t + 1) ‒ 3ct = 4
3c = 4
3c =4
c =4/3
4
∴ P.I. = t.(3) t
3
t +
t.(3)t
&
∴ The general solution is yt = A.(3) +

Example 12 Solve yt +1 + 7 yt = 2t 5t

Solution In this case C.F . = A(−7)t and we choose yt = ( p0 + p1t )5t as our trial
function for finding P.I.because φ (t ) is the product ofpolynomial of
degree 1 and 5t and 5 is not a root of the associated characteristic
equation we have.
Now, from yt, we can ascertain yt+1: yt+1 = [p0 + p1(t +1)]5t+1

288
 Substituting values of yt and yt+1 in the given difference equation Linear Difference
Equations
yt +1 + 7 yt = 2t 5t , we get [p0 + p1(t +1)]5t+1+ 7(p0 + p1t )5t = 2t5t

Dividing both sides by 5t, we get

5 p0 + 5 p1t + 5 p1 + 7 p0 + 7 p1t = 2t
12 p0 + 5 p1 + 12 p1t = 2t
On comparing coefficients of like terms, we get
12 p0 + 5 p1 = 0 and 12 p1 = 2
5
⇒ p1 = and p0 = − 72
(
5 1
∴ P.I. is given by: yp = (− 72 + 6 /)5t
5 1
Thus, the general solution is yt = A ( −7)t+ (− 72 + 6 /)5
t

Check Your Progress 2

1) How will you identify if a difference equation is a second-order one?
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
2) Solve the following first-order linear difference equations:
i) yt+1 + 2yt = 5
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
ii) yt +1 + 7 yt = 1 + t + t 2

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
iii) yt +1 + yt = 7t

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
iv) −4 = 5(4)
………………………………………………………………………
………………………………………………………………………
289
Difference Equations ………………………………………………………………………
………………………………………………………………………
3) Solve the following second-order linear difference equations:
i) yt +2 − 5 yt +1 + 6 yt = 0

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
ii) yt +2 + 4 yt +1 + 4 yt = 0

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
iii) yt+2− 11yt+1 + 28yt = 6
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
iv) yt +2 − 3 yt +1 − 40 yt = 1 + t

………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
v) yt+2− 6yt+1 + 9yt = t + t2
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………

15.5 ECONOMIC APPLICATIONS

Difference equations have wide ranging applications in economics. In dynamic
analysis, when the time variable t is allowed to take only discrete values and
we are interested in finding a time path, given the pattern of change of a
variable y over time variable t, the difference equations come out to be quite
handy. We take up some situations in economics where the problem can be
converted into a difference equation whose solution provides a solution to the
problem in hand.
15.5.1 The Cobweb Model
Consider a situation in economic analysis where the demand function and the
supply functions for a given commodity are given by
Qst = S ( pt ) = −c + dpt −1 and Qdt = D( pt ) = a − bpt , respectively.
290
Qst is the quantity supplied at time t and Qdt the quantity demanded at time t. Linear Difference
Equations
pt denotes the price of the commodity at time t. Constants a, b, c, and d are all
taken to be positive, , with b representing the demand sensitivity to price at
time t and d representing supply sensitivity to price at time t – 1, respectively.
In the supply function equation, the quantity to be supplied by the producer has
been expressed in terms of the price in the time period t − 1 . Such situations do
occur in Economics. For example, a producer may base his future output
decision on the current price of the commodity (i.e., Qt+1 may depend on pt), or
the current output decision on the price of the commodity in a previous time
period (i.e., Qt may depend on pt−1).
Assume that in each time period, the market price is always set at a level such
that quantity demanded is equal to the quantity supplied. That is,
Qst = Qdt
The equations and the assumptions considered above, give us a market model
for a single commodity. Substituting the values of Qst and Qdt in the above
equation, we get
−c + dpt −1 = a − bpt

⇒ bpt + dpt −1 = a + c

On replacing t by t+1 in the above equation, we obtain

⇒ bpt+1 + dpt = a + c
2 4
⇒pt+1 + pt =
3 3
…(xv)

Note that the equation (xv) is nothing but a linear non-homogeneous difference
equation of order 1. Thus, we have seen how a situation in Economics gets
translated into a difference equation.
Let us now solve Equation (xv). The corresponding homogenous difference
equation is
2
pt+1 + pt= 0
3

and the characteristic equation (with pt+1 = x and pt= 1) is

d
x+ =0
b
d
The only root of the characteristic equation is −
b
d
∴ C .F . = A( − )t
b
whereA is a constant. To find P.I., we observe that in equation (xv) φ (t ) =
a+c
, which is a constant. Thus we choose pt= k as our trial equation. Also,
b
pt+1= k. On substituting pt and pt+1 in (xv), we get
d a+c
k+ k=
b b
291
 4
⇒
Difference Equations
3 2
k=
Thus, the general solution of equation (xv) is
5 4
6
)
3 2
pt= A(− + …(xvi)

If we are given some initial condition, say p0i.e.,the price at t = 0, the value of
the constant A can be obtained. On substituting t = 0 in the general solution
equation, we get
4 7+8
p0 = A + ⇒ A = p0 − 6+5
3 2
Now, equation (xvi) can be rewritten as

5 4
7+8
pt= (p0 − 6+5) (−
6
)
3 2
+ …(xvii)

Equation (xvii), which is a solution to the difference equation (xv), represents

the time path for pt. That is, given t we can find the corresponding equilibrium
price pt. This gives an oscillatory time path (a regular back and forth path) and
the model is known as the cobweb model. There are three varieties of
oscillation patterns in the model. The model is called explosive if d>b,
uniform if d = b and damped if d<b. The model is called the cobweb model
because when we trace out the prices and the quantities in subsequent periods,
a cobweb is spun around the demand and supply curves.
The cobweb model was an example of a situation in Economics which gives
rise to a first order difference equation. We now consider a situation which
gives rise to a second order difference equation.
15.5.2 Samuelson Multiplier-Accelerator Interaction Model
This model assumes that the national income yt consists of three components
i) Consumption (Ct)
ii) Investment (It)
iii) Government Expenditure (Gt)
The Consumption (Ct) in the time period t is assumed to be proportional to the
income of the previous period, yt−1.
Investment (It) in period t is assumed to be proportional to the increase in the
consumption in the time period t over the consumption in the time period t –1,
i.e., It is assumed to be proportional to Ct−C t −1. These set of assumptions give
rise to the following set of equations:
Income Function yt = Ct+ It+ G0
Consumption FunctionCt = γyt−−1 (0 <γ< 1)
Investment Function It = α(Ct−C t −1) (α> 0)
The government expenditure is assumed to be constant equal to G0 .The
constant γrepresents the marginal propensity to consume and α represents the
acceleration coefficient.
292
Substituting the value Ct from consumption function into investment function, Linear Difference
we get Equations

I t = αγ ( yt −1 − yt −2 )

Now substituting the value of consumption function and the value of Itobtained
above in the income function (yt), we get
yt = γyt−−1+αγ(yt−−1−yt−−2)+ G0
⇒ yt−γyt−−1 −αγ(yt−−1−yt−−2) = G0
⇒ yt−γ (1 + α)yt−−1 + αγyt−−2 = G0
By replacing t by t + 2 in the above equation, this equation can be rewritten as
⇒ yt+2−γ (1 + α) yt+1 + αγyt = G0
Now we have a linear non-homogeneous difference equation of order 2 that
can be easily solved.
Since φ (t ) = G0 , a constant, the particular integral (P.I) is easy to obtain. The
calculation of complimentary function (C.F) may pose some problems because
in this case the characteristic equation of the associated homogeneous
difference equation (with yt+2 = x2, yt+1 = x, and yt= 1) is
x2−γ (1 + α) x+ αγ= 0
This equation has roots

9( ) ± :92 ( ); $+ 9
x=

−6 ± √6 − 478
<∵ = = for a quadratic equation, 7= + 6= + 8 = 0K
27

The nature of roots depends on the constants α and γ,and a number of cases
may arise. We have no intention of going into the complicated details as we
have already established our claim that there are situations in economic
analysis which can be tackled using difference equations.
Check Your Progress 3
1) Obtain the difference equation in each of the following special cases of
the cobweb model and solve it
i) Qdt = 18 − 3Pt , Qst = −3 + 4 Pt −1

ii) Qdt = 5 − 2 Pt , Qst = −2 + 5Pt −1

iii) Qdt = 19 − 7 Pt Qst = 6 Pt −1 − 4

whereQdt stands for quantity demanded, Qst stands for quantity supplied
and P is the price.
…………………………………………………………………………….
…………………………………………………………………………….
……………………………………………………………………………. 293
Difference Equations …………………………………………………………………………….
…………………………………………………………………………….
2) Obtain the difference equation in each of the following special cases of
the Samuelson multiplier accelerator interaction model given by the
income, consumption and investment function equations in section 15.5.2
and solve them.
2 49
i) Take α = , γ =
3 50
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….
2 24
ii) Take α = , γ =
3 25
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….
……………………………………………………………………….

15.6 LET US SUM UP

This unit dealt with dynamic economic processes in discrete time. The next
unit too will deal with discrete dynamic processes, but the processes studied
will be non-linear. The present unit began by explaining the difference between
continuous and discrete processes. After touching upon first-forward
difference, the unit went on to explain the nature of a difference equation. You
were familiarised with homogeneous and non-homogeneous difference
equation.
Subsequently, the unit discussed first order and second order difference
equations. The unit considered linear equations only. Non-linear difference
equations would have taken us beyond the scope of this unit. The unit
explained in detail the method of solving first order and second order
difference equations. The iterative method as well as the general method of
solving first order difference equations were explained. The method of solving
second order difference equations was discussed, and several examples were
provided. Finally, the unit discussed the use of difference equation in economic
modeling, with applications to the Cobweb model and the Samuelson
multiplier-accelerator interaction model.

294
 Linear Difference
15.7 ANSWERS/HINTS TO CHECK YOUR Equations
PROGRESS EXERCISES
Check Your Progress 1
1) A difference equation is homogeneous if its constant term, the term that
does not contains y (that is the variable) is zero, otherwise it is non-
homogenous.
2) i) =6 ,

ii) = ,

iii) = 10/ + ,

i) =3 ,

Check Your Progress 2

1) A difference equation that does not contain successive values of yt greater
than yt+2is said to be of second-order.
%
2) i) = (3) + &
% &
ii) = (−7) + L %(
+ & / + */ M
N N
iii) = (−1) − + + /
%
iv) = (4) + + /(4)

3) i) = (2) + (3)
ii) =( +/ )(−2)

iii) = (4) + (7) +

&

+
iv) = (8) + (−5) + L− − + /M
N(+

N &
v) =( +/ )(3) + L* + + / + + / M

Check Your Progress 3

+
1) i) P = (− &) + 3
%
ii) P = (− ) + 1
( &
iii) P = (− N) + &

+ N
2) i) = %
+ ,
+ 50 Q,

+
ii) = %
+ 25 Q,

295
Difference Equations
UNIT 16 NON-LINEAR DIFFERENCE
EQUATIONS*
Structure
16.0 Objectives
16.1 Introduction
16.2 Phase Diagram and Qualitative Analysis
16.3 Linearising Non-Linear Difference Equations
16.4 Applications of Non-Linear Difference Equations
16.4.1 Solow Growth Model
16.4.2 Cycles and Chaos
16.5 Let Us Sum Up
16.6 Answers / Hints to Check Your Progress Exercises

16.0 OBJECTIVES
After going through this Unit, you will be able to:
• Differentiate between linear and non-linear difference equations;
• Explain the concept of a phase diagram and the important role that phase
diagrams play in understanding non-linear difference equations;
• Discuss how qualitative analysis of non-linear difference equations is
carried out;
• Describe the procedure for linearising non-linear difference equations;
and
• Discuss the application of non-linear difference equation to the Solow
growth model and to Cycles and Chaos theory.

16.1 INTRODUCTION
In the previous unit, we had discussed linear difference equations. In practice,
several economic models give rise to non-linear dynamic relations. The
introduction of non-linearity does not alter the essence of a difference
equation. Let us consider a simple economic system. Suppose that the
economic system under discussion can be described by a single variable
x ∈ X , where X⊆ ℝ is a non-empty interval of the real number line. You
have studied about real number line and subsets in the unit on sets. The
variable x is referred to as the system variable and X is referred to as the
system domain. The system domain consists of all possible values of the
system variable. In our context, we think of the economic system as a
dynamic one, that is, that the values of x change over time. We assume that
time is measured in discrete periods ∈ {0,1,2,3. . }. The variable x changes
only once in one period. The set Z+ (i.e. a set of positive integers)is the set of
all time periods and is called the time domain or time horizon. Time is

296 * Contributed by Shri Saugato Sen & Chetali Arora

indicated by a subscript on the variable x. Thus xt is the value of x in time Non-Linear Difference
period t. All difference equations describe the evolution of a variable over Equations
time. The dynamic evolution of the system is described by a function f, which
shows how xt is dependent on past values of x. We write
=
Here the function f is called the law of motion or the system dynamics. In the
previous unit, we were concerned with difference equations that could be
described by a linear equation like
= +

The non-linear form includes the linear form as a special case, but has the
advantage of allowing a much broader range of varieties of time paths to
emerge.

Actually the non-linear difference equation of the above type can be expressed
as xt +1 = f ( xt , t ) . Here t itself appears as an argument, that is, an independent
variable. However, we consider only autonomous difference equation where
time does not appear as an independent variable. Hence we restrict ourselves
to the study of equations of the type

=
Notice that xt+1 depends only on the value of x in the period immediately
preceding to it, that is, on xt and not on and other past values of x. Thus
the value of x in any period depends only on the value of x in the immediately
preceding period and not on the values of x in distant past periods. This type of
difference equation is called a first-order difference equation. In this unit, we
will only consider first-order difference equation.
A general form of linear difference equation would be
= , , ,…, ,
wherek is a positive integer called the order of the difference equation. The
above equation also considers time as an independent variable. As we saw
above, if t does not enter as a variable itself, then the difference equation is
autonomous.
The solution to the autonomous difference equation of degree one, =
, is a sequence { xt }t = 0 , such that xt ∈ X and the equation =
∞

holds. Such a sequence is referred to as the trajectory of the equation

= .
Let us look again at our difference equation, = . Notice that we can
shift it by a period and write xt = f ( xt −1 ) .

Given an initial condition x0 at t = 0, we can work out a solution sequence as

follows:

x1 = f ( x0 )

x2 = f ( x1 ) = f  f ( x0 )  = f 2 ( x 0 )

x3 = f ( x2 ) = f  f 2 ( x0 )  = f 3 ( x0 )
297
Difference Equations .
.
.
.

xt = f ( xt −1 ) = f  f t −1 ( x0 )  = f t ( x0 )

Where, f t denotes the t th iterate of f, not its power. The function

x ( t , x0 ) = f t ( x0 ) is called the flow of the system xt = f ( xt −1 )

There is very simple but powerful method of analysing the dynamics of the
above equation. To use this technique, we must draw the graph of the function
f, as well as a 45o line into a { xt , xt +1} plane. We take up the discussion of
trying to arrive at a solution of first-order difference equation by making use
of such diagrams (called phase diagrams) in the next section.
After that, the unit will discuss how non-linear difference equations can be
approximated linearly in small neighbourhoods around specific points.
Following this the unit will discuss a few applications of non-linear difference
equations in section 16.4. First, the unit will discuss the growth model
propounded by Solow. Finally the unit will discuss periodic cycles as well as
aperiodic behavior of economic systems. This aperiodic behavior is called
chaos.

16.2 PHASE DIAGRAMS AND QUALITATIVE

ANALYSIS
We draw phase diagrams in the case of non-linear equations to help us get a
qualitative ‘solution’ of a difference equation. To understand phase diagrams,
consider a difference equation
= *
, where t = 0, 1, 2, 3…
Case I: 0 < ′ < 1 and <0
We first consider the phase diagrams of the non-linear difference equation by
assuming that the function is upward sloping, that is > 0 and that the
function flattens as x increases, that is, < 0.

*A general non-linear, first order difference equation is given by = , ,t =

0, 1, 2…Here, we are considering an equation which does not explicitly depends on
time, that is = , where t = 0, 1, 2, 3…This is called nonlinear, first-
order, autonomous difference equation.
If we plot this curve on a graph which has xt+1 on the vertical axis and xton the
horizontal axis, we get a picture as in Figure16.1. Let us denote this curve (also
called the phase line) equation by
= ! …(1)
We have also plotted a 45° line in diagram 16.1. The intersection of the curve
fawith the 45°0 line marks an equilibrium point where xt +1 = xt . Given an initial
value x0 on the horizontal axis, from our equation 1, we get the vertical-
coordinate value as x1 [where = ! # ], this is what is mapped by the
Phase line fa in figure 16.1; considering the other way round, x0 on the
298
horizontal axis, marked straight up to the phase line at point A gives the value Non-Linear Difference
of x1. Next we map another pair given by x1 and x2 [where, = ! ]. Note Equations

that x1is transplotted from vertical axis to the horizontal axis with the help of
45° line (having slope = 1). Steps are repeated in similar fashion to plot
subsequent pairs. This process is what we learnt in previous unit, called the
Iteration method.

%& + 1

T
+,

A
B

45#
o %&
# ̅

′′
Figure 16.1: Phase Diagram for 0 < ′ < 1 and <0

On similar lines, the Graphic iterations discussed above will be applicable to

the following three cases as well:
Case II: ′ > 1 and <0
Let Phase line equation be give by = $

+-
%& + 1

45#

o ̅ # %&

′′
Figure 16.2: Phase Diagram for ′ > 1 and <0

299
Difference Equations Case III: −1 < ′ <0
Let Phase line equation be give by = /

%& + 1

+1

45#

o # ̅ %&

Figure 16.3: Phase Diagram for −1 < ′ <0

Case IV: ′ < −1

Let Phase line equation be give by = 0

%& + 1 +2

45#

o %&
# ̅

Figure 16.4: Phase Diagram for ′ < −1

The above four cases illustrate four basic types of Phase lines, each indicating a
different time path. The intertemporal (where ‘intertemporal’ refers to ‘across
the time’) equilibrium value of x denoted by ̅ is given by the intersection of
the phase lines in each diagram with the respective 45° line, labeled as point T.
At point T, =
300
Our next task is to see whether given an initial value x0 ≠ ̅ , the pattern of Non-Linear Difference
change as indicated by the phase line will lead consistently towards ̅ — which Equations

is the case of Convergence, or away from it — the case of Divergence. In Case

I above, phase line fa leads us from x0 to ̅ in a steady path, without oscillation.
Even when x0 is placed to the right of ̅ , the movement towards ̅ will be
steady and in the left direction. This is referred to as convergence to the
equilibrium. Under Case II, phase line fb with a slope exceeding 1, showcases a
divergent path. Here beginning from an initial value x0 greater than ̅ , we are
lead steadily away from the equilibrium value to higher and higher values of x.
An initial value, lower than ̅ , will give rise to similar steady divergent
movements in the opposite direction.
In Cases III and IV, we encounter steady oscillatory movements, along with the
phenomenon of overshooting the equilibrium mark. Phase line fc with absolute
slope less than 1, will result in convergence, along with the extent of
overshooting (given by x0 leading to x1 which exceeds ̅ to be followed by x2
falling short of ̅ ) diminishing in successive periods. In case of phase line fd
whose absolute slope exceeds 1, there is an opposite scenario of a divergent
time path.
Thus, two basic rules can be drawn from above discussion:
i) A first-order, autonomous, non-linear difference equation will have a
locally stable steady-state equilibrium point ( ̅ ) if the absolute value of
the derivative (which is nothing but the slope of the phase line), given by
′ ̅ is less than 1, whereas will be unstable if the absolute value of the
derivative is greater than 1 at that point (i.e. at ̅ ).
ii) A first-order, autonomous, non-linear difference equation will lead to
oscillations in xt if ′ is negative for all xt> 0, whereas there will be
monotonic movement in xt if ′ is positive for all xt> 0.
Check Your Progress 1
1) What is a non-linear difference equation?
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
2) What do you understand by a phase diagram? What is it used for?
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
……………………………………………………………………………. 301
Difference Equations 3) What is the basic difference in the equilibrium of a linear difference
equation and that of a non-linear one?
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….

16.3 LINEARISING NON-LINEAR DIFFERENCE

EQUATIONS
After obtaining equilibrium point(s), that is, the fixed point(s) of a difference
equation, an important task left is to explore the stability of that equilibrium.
For non-linear equations, we can only explore stability of equilibrium point(s)
in a small neighbourhood. This is done by linearising the non-linear function
at equilibrium and then testing the slope of the linear approximation. In fact,
this is just an extension of what we did when we studied phase diagrams in the
previous section: we looked at the slope of f (x) function in the phase diagram
to judge about the stability of the equilibrium. We investigated the slope of f
(x) in the region of the equilibrium.

You have seen in an earlier unit how we use Taylor’s series expansion to
linearise a non-linear function. Here too, we shall use Taylor’s series
expansion. We attempt to linearisef (x), i.e. our Non-linear difference function
by finding a first-order Taylor series expansion of this function, with
equilibrium as the point of expansion. Although, a first-order Taylor series
expansion produces a linear approximation to a non-linear function, the
problem is that the approximation is good only for a limited range around the
point of expansion. Also, the greater the degree of curvature of the original
function, the smaller that range will be. Of course, our expansion need not be
limited to first-order only, and we can take the expansion to as high an order
as we like. Moreover, if the original curve is very greatly curved, we may
require higher orders of expansion to approximate the original curve, but the
higher orders of expansion introduce non-linear elements into the expansion,
and since the idea is to take a linear expansion, we stop at the first-order, that
is, at a linear expansion.

To take a first-order Taylor series approximation to a general function f (x),

we begin by choosing the value of x which would determine the point around
which we will construct a linear approximation to the non-linear function. Let
us denote this value of x by x*,which means that the value of the function f (x)
at the point of approximation is f (x*). Then we can write, as the
approximation to the function f (x) at some arbitrary point x*:

df ( x* )
f ( x ) ≈ f ( x* ) + ( x − x* )
dx

Note that the derivative on the right-hand side of the equation is also evaluated
at x*. The closer x is to x*, the closer the value of the approximation.

302
Now we apply this approximation to a first-order difference equation. Non-Linear Difference
Remember we are using the following equation as our first-order difference Equations
equation:
=
and using x* or ̅ to denote an equilibrium of the system. Approximating the
function close to the equilibrium gives

df ( x* )
xt +1 = f ( xt ) = f ( x ) + *
(x − x )
t
*

dx

Now note that since x*is an equilibrium point, f ( x* ) = x* . Hence we can

*
df ( x* )
write the above equation as xt +1 = x + ( xt − x* ) …(2)
dx
Let us define a new variable xD, as the deviation of the current value of x from
its equilibrium value x*. Thus,xDt = xt– x* and xDt+1 = xt+1 – x*. Hence, we can
write the equation (2) as

D
df ( x* )
xt +1 = xtD
dx
…(3)
In interpreting equation 3, we must remember that we have evaluated the first
df ( x* )
derivative at a single point (here the equilibrium point), thus it is a
dx
constant. Given this, equation 3 becomes a homogeneous equation in xD, with
constant coefficient. This makes it a linear first-order homogeneous difference
equation. Since it is a homogeneous equation, it means that its equilibrium is
at xD = 0, but since xDis the deviation of the original untransformed variable
from its equilibrium, it means when xD = 0, x = x*. So if equation 2 is stable
with xD converging to its equilibrium, then x too must converge to its own
equilibrium.
Check Your Progress 2
1) Why do we need to linearise non-linear difference equations?
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
2) Suggest a method by which we can approximate a first-order non-linear
difference equation near a small neighbourhood of its equilibrium point.
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
303
Difference Equations
16.4 APPLICATIONS OF NON-LINEAR
DIFFERENCE EQUATIONS
In this section we discuss some applications of non-linear difference
equations. Probably the most familiar among the various applications are some
models of economic growth. Even several consumption models display non-
linear relations, especially when the utility function is not very restrictive.
However we are not discussing consumption functions here, but taking up
other non-linear dynamic processes. Specifically we shall take up for
discussion a non-linear growth model suggested by Robert Solow, and Cycles
and Chaos.
16.4.1 The Solow Model
Let us see what we understand by a growth model. Since we are discussing
dynamics, we would want to see how a variable evolves over time. Here time
is expected to pass in discrete levels. In the theory of economic growth, we see
how the aggregate economic output or income grows over time. Your
“Principles of Microeconomics” course in this semester will introduce you to
the concept of a production function, where output is a function of labour and
capital, in your. Here, we will think of an aggregate production function for
the economy as a whole. Your “Principles of Macroeconomics” course will
introduce you to the study of the economy in aggregate, that is, the
Macroeconomics aspect of an economy. By combining our understanding
from both the courses, here we are basically examining how the aggregate
output or what we call the Gross Domestic Product (GDP) of an economy
grows as a result of growth of total labour and capital (machines and
equipment) over time.
Professor Robert Solow in 1956 propounded a model of economic growth.
This model has become the central growth model in economic analysis, and
has served as the baseline growth model upon which further work on growth
has been carried out and new models suggested. Prof. Solow went on to
receive the Nobel Prize in Economics in 1987 for this model. The economic
growth model of Robert Solow which we consider here is also known as the
Neoclassical growth model. This model contains difference equations for two
variables, but by a trick common to growth models we are able to reduce it to
a single difference equation model.
We begin with an aggregate production function:
3 = 4 ,5
Where Y is aggregate output, K is aggregate capital and L is aggregate labour.
The time subscripts on each variable indicate that there are no lags in the
production process.
For simplicity, labour (which is here assumed to be identical to population;
that is, the labour force participation rate is100 per cent) is assumed to grow at
an exogenous proportional rate n. In other words, we get a difference
equation:
5 = 1+6 5
Capital grows as a result of net investment, which is given by gross
investment minus an allowance for depreciation. Thus, Net investment is
defined as the change in the stock of capital in two successive time periods.
304
This is a neoclassical model, so all savings are invested in productive physical Non-Linear Difference
capital, hence investments are considered to be equal to the savings. Consider Equations
the following equation that gives the value of capital at period t (Kt) as a
function of capital at period t – 1 (K t – 1):

Kt = sF ( Kt −1 , Lt −1 ) + (1 − δ ) Kt −1
7
Here, s is average total saving per output, . So S = sY = sF ( K , L ) . We are
8
also considering saving S equals investment I. So we can write, I = sF ( K , L ) .
It must be clear that,
I t −1 = Kt − Kt −1. This means Kt = I t −1 + Kt −1. If we subtract depreciation at rate δ ,
we get
Kt = I t −1 + (1 − δ ) Kt −1. Substituting for I t −1 as sF ( K t −1 , Lt −1 ) , we can write
Kt = sF ( Kt −1 , Lt −1 ) + (1 − δ ) Kt −1

Note that saving is done in period t – 1 for new capital to emerge in period t.
The above equation tells us that the current period’s capital (Kt) is equal to the
part of last period’s capital left after deducting for depreciation [(1 − 9 4 ]
plus saving (equaling investment) done out of last year’s income (output)
which shows up in new capital goods and equipment in the current period, i.e.,
[;< 4 , 5 ].
Now let us make an assumption about the production function. Let us assume
that the production function displays Constant Returns to Scale (CRS). This
means that if labour and capital are scaled up by a factor λ, the output will
increase by the same factor λ.
That is, CRS implies >3 = < >4 , >5

F ( K t , Lt ) K 
Let us assume λ = . Then, = F  t ,1
?@ Lt  Lt 
Kt
where is the current capital-labour ratio. Right-hand side of the equation
Lt
shows the amount a single worker would produce if he or she had available a
capital stock equal to the current capital-labour ratio. The left-hand side of this
equation is output per worker and is denoted yt . We can write the above
K
equation as yt = f (kt ) where, k t = t .Here, we have depicted output per
Lt
capita as a function of capital per capita.
Recall the equation K t = sF (K t −1 , Lt −1 ) + (1 − δ )K t −1 . Let us divide both sides of
Kt F (K t −1 , Lt −1 ) (1 − δ )K t −1
the above equation by Lt. We get =s +
Lt Lt Lt

In this equation, the right hand side has terms pertaining to both time periods
t −1 and t. To resolve this, let us multiply and divide all terms on the right
hand side by 5 . We get

K t  sF (K t −1 , Lt −1 )  Lt −1   (1 − δ )K t −1  Lt −1 
=   +    …(4)
Lt  Lt −1  Lt   Lt −1  Lt 
305
Difference Equations F (K t −1 , Lt −1 )
The term in above equation is output per worker in time period
Lt −1
K
t −1 and the term t −1 is capital-labour ratio in period t −1. We hadseen
Lt −1
L 1
above that Lt = (1 + n )Lt −1 . From this we get t −1 = . Using the notation
Lt (1 + n )
for per capita quantities we had developed earlier, we can write equation (4)
sf (k t −1 )  1 − δ 
as, kt = + k t −1 …(5)
1+ n 1+ n 
Here n and δ are exogenous (i.e., given and fixed from outside the model).
Hence, we get a first-order difference equation given by equation 5, where A
is a function of A . We have not specified a precise functional form for f
(A ), so we limit ourselves to a qualitative, phase diagram analysis of equation
(5). We can see that per capita production function yt = f (kt ) has usual
production function properties. Also f ′(k ) > 0, f ′′(k ) < 0. considering these
assumptions, we can draw a phase diagram of the type given in figure 16.5.

B&

B& = + B&−C
E
A3
A2

A1

45°
B& C
0 A0 A1 A2 A∗

Figure 16.5: Phase Diagram for a Neoclassical Growth Model

At equilibrium (that is at point E), kt −1 = kt = k * , say. Then, even though we

may not have a precise definition of f (k), we can see that equation (B) must
satisfy the relationship

( )
f k*
=
(n + δ )
*
k s
16.4.2 Cycles and Chaos
Till now we have considered only those non-linear difference equations for
which the slope of the function f (.), does not change sign, that is, the graph of
xt+1 against xt, or xt against xt-1, does not change sign. In other words, the graph
306 of xt against xt-1 in phase diagram, is either monotonically increasing or
monotonically decreasing, but never has the shape of a hill (inverted U), or Non-Linear Difference
Equations
valley (U-shaped). In this sub-section, we consider non-linear difference
equations that generate hill-shaped curves in the phase diagram. This kind of
difference equation generates interesting dynamic behavior, such as Cycles
which repeat themselves every two or three periods, or even dynamic
processes in which there is irregularity in the behavior of xt. This type of an
irregular process is called Chaos. It is beyond the scope of this unit to provide
a detailed analysis of such cycles as well as chaotic behavior. In this sub-
section, we give a simple exposition of the topic.
Consider the first-order, nonlinear, autonomous difference equation
xt +1 = A xt(1− xt), where t = 0, 1, 2,... …(6)
The equilibrium values ̅ are obtained by solving
̅−F ̅ 1− ̅ =0⟹F ̅ −F ̅+ ̅ =0
1−F
⟹F ̅ + ̅ 1−F = 0 ⟹ ̅ + ̅ =0
F
J
This gives ̅ HI K + ̅L = 0
J
J
The two steady points are ̅ = 0and ̅=I J
K.

From the second steady-state point, we can say that a strictly positive steady-
state equilibrium exists only if A> 1. If A ≤ 1, then the steady states are zero or
negative. These we do not discuss as they do not have much relevance in
Economics. Recall from our discussion of phase diagrams that a steady-state
equilibrium point of any first-order autonomous, non-linear difference equation
is locally stable, if the absolute value of its slope, that is its derivative is less
than 1at that point. The value of the derivative of equation 6 above at the two
J
steady points ̅ = 0 and ̅ = I J K we found above is given as follows:

M
= F − 2F
M
0N@OP J 0N@OP
At ̅ = 0, 0N@
= F and at ̅ = J
, 0N@
=2−F

The above result implies that the steady point ̅ = 0 is unstable (as we assumed
J
A> 1). The point ̅ = J will be locally stable only if absolute value of 2 − F,
that is |2 − F|< 1⟹ 1 <A < 3. This was what concluded the basic rule
(i) under section 16.2.

Since the above equation is a first-order nonlinear difference equation, we can

draw a phase diagram for the equation. The intersection of the phase line with
J
the 45° line will occur at our steady points, that is at 0and I J K. The graph
0N@OP
will peak at x= , where the slope given by 0N@
= 0. Second derivative of the
0R N@OP
function, given by = −2F, which is negative, implying that the graph of
0N@ R
the equation will be inverted U shape. (refer Figure 16.6)

307
Difference Equations
%& C

45°
%&
0 # ̅

Figure 16.6: Phase Diagram for the first-order, nonlinear, autonomous difference
equation xt +1 = Axt(1−xt).

Note yourself from the graph that the intersection of the phase line with the 45°
line at point will happen to the left of the peak, if 1 <A < 2. This implies that
J
the slope is positive at the stable-steady point ̅ = J . Whereas, when 2 <A<
3, the intersection will be on the right of the steady point. This satisfies the
condition of local stability, as the slope of the phase line will be negative at the
stable steady-state point.

A negative slope with less than 1 absolute value will mean that xt will converge
to ̅ from either direction within a neighbourhood, but the path of approaching
it will be oscillatory. Refer the figure. Starting from x0, the slope is positive,
with xt increasing monotonically in the initial few periods. However as xt
approaches the neighbourhoodof ̅ , the slope becomes negative with xt
becoming oscillating in the neighbourhood before it converges to the steady
state.
J
What will be the behavior of the phase line when A ≥ 3? Firstly, ̅ = J will
no longer be a stable steady state, rather will be unstable. Secondly, the hill-
shaped phase line possesses a peculiar characteristic which does not hold for a
monotonically phase diagram—that is, xt will not diverge endlessly to 0 or
infinity, but will be oscillating within a bound range, though it will not be
converging to the steady state, but could converge to regular periodic behavior.

When a non-linear difference equation throws up this kind of a inverted U-

shape phase diagram, there arise threshold in the behavior of the function in
the sense that small changes in the value of A can lead to dramatic changes in
the behavior of x and in its trajectory. For instance, in our above example, if
the value of A lies between 2 and 3, x follows a simple convergent
alternations, that is, x takes on alternate values but moves towards
convergence. As the value of A becomes 3 or above, the trajectories become
very complicated. For some values between 3 and 4, xt settles into periodic
308 alternations, or what is called a limit cycle. This basically means that x takes
values only within a certain range, and alternates values within this range. It Non-Linear Difference
Equations
can be shown that when the value of A is 3.2, if the dynamic system runs long
enough, the system will settle down to a pattern that is called a period 2 cycle,
getting back and forth between a value of 0.513 and 0.799. This is an example
of the alternation version of the limit cycle. Limit cycles produce oscillations
but these appear mainly in the case of higher order difference equations. The
alternations, however, give the essence of a limit cycle. This is an example of
a stable limit cycle.

However, there can be cases when the limit cycles are unstable. These cycles
have the same basic properties as an unstable equilibrium. The basic feature is
that if we start from a value in the limit cycle, we will stay in that same
cyclical path always, neither converging nor diverging; if we start from a
value on either side of the cycle, we will diverge from it.

If we increase the value of A steadily, different results emerge. At A = 3.4, the

range of the interval within which the value of x lies is bigger, but it is still
between two values. If A = 3.5, the system follows a period 4 cycle, that is
going from x = 0.382 to x = 0.827 to x = 0.501 to x = 0.875. but if A is slightly
higher at A = 3.84, the system is back to a period three cycle.

The interesting thing about values of A, and this is where chaos enters, is that
the periodicity of alternations is not always smooth. If we set A = 3.58, we
will find that the system alternates around an upper equilibrium, but never
reapeats itself. It does not display any pattern which repeats over and over. It
becomes aperiodic. In other words, the system is chaotic.

In what way is the study of chaos useful in economics? First of all, chaos is
very hard to distinguish from a random process. And what is a random
process? You will study about uncertainty and probability in the course on
Statistics in the next semester. Some of you may already be familiar with
probability. Random processes are associated with probability. A random
variable is a variable whose value is determined by chance. Non-random or
non-probability-based processes are called deterministic. The dynamic
processes that we have been studying using non-linear (or even linear in the
previous unit) difference equation is a deterministic process. The significance
of chaos is that chaos depicts a situation where a deterministic process mimics
the pattern of a random process. It is very difficult to distinguish a random
process from a chaotic process. A fundamental difference between a random
process and a chaotic process is that a random process cannot be predicted.
However in a chaotic system, given the parameters, future values can fairly
easily be predicted.
Check Your Progress 3

1) Explain the fundamental difference equation in the Solow growth model.

…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
309
Difference Equations
2) a) What is the basic shape of the phase diagram of a chaotic system.
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
b) What do you understand by a limit cycle?
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
………………………………………………………………………
3) How is chaos important in economics? In what way is a chaotic process
different from a random process?
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….
…………………………………………………………………………….

16. 5 LET US SUM UP

This unit followed unit 15 which was on linear difference equation. Like the
previous unit, this unit dealt with discrete dynamic processes, that is, processes
where the value of a variable depends on the past values of the same variable.
The variables change values in discrete steps rather than continuously. The
unit was concerned with general difference equations, of which linear
difference equations were of one special type. Hence, the term ‘non-linear’ in
the title is to be interpreted as general difference equations, rather than being
restricted to the case of difference equations that are not linear.

The Unit began with explaining the general nature of difference equations. We
saw that linear difference equations are a specific type of difference equations.
The unit restricted itself to first-order difference equations, that is, to equations
in which the value of the variable in period t depends only on value of the
variable in period t −1, and not on the values in time periods t −2,t −3, and so
on. The value of the variable thus depends on the immediate past value of the
variable and not on the values of the variable in the distant past. We saw that
we can shift the time period of the equation and still maintain the first order of
the equation by considering that the value of the variable in period t + 1
depends only on the value of the variable in period t.
310
In the next section, the Unit introduced you to phase diagrams (or phase line) Non-Linear Difference
as a way to depict first-order, nonlinear, autonomous difference equations, and Equations
say something about its possible solutions. The phase diagram is constructed
by showing the current value of the variable on the horizontal axis and the
future value of the variable on the vertical axis; or depicting the past value of
the variable on the horizontal axis and the current value on the vertical axis.
We saw that unlike linear difference equations of the first degree, non-linear
difference equations of the first degree can intersect the 45° line at more than
one place. If the curve depicting the difference equation is upward sloping
then the equilibrium point (fixed point) can be converging (attractor) or
diverging (repellor). A downward sloping curve leads to oscillations. Since the
unit restricted itself to first-order difference equations, the unit relied mainly
on graphic solution methods and hence phase diagrams became a very
important tool of analysis.

In the next section, we took a look at linearising non-linear difference

equations. To this end, we utilised first-order Taylor Series expansion of the
function f (x) at the relevant point x*. In the subsequent section, the unit took
up for discussion two applications of non-linear difference equations: the
Solow growth model, and a simple look at periodic Cycles as well as periodic
Chaos. The Solow growth model showed the dynamics of how output per
worker evolves over time as a function of capital per worker. A suitable phase
diagram was provided with capital-worker ratio in one period shown on the
vertical axis, and output-worker ratio shown on the horizontal axis. We
showed the dynamics with the help of a non-linear difference equation relating
capital per worker of the current period to the capital per worker of the
previous period.

Finally, the unit discussed dynamic processes where the phase diagram of the
non-linear function is inverted U-shaped. Important features and properties of
such phase diagrams were discussed depending on whether the maximum
point of the function lay to the left of the 450 line, or to its right. From such a
quadratic difference equation, we studied the properties of limit cycles, as well
as aperiodic cycles. The latter situation is called chaos, and at the end we saw
that chaotic behavior mimics the behavior of random processes.

16.6 ANSWERS/HINTS TO CHECK YOUR

PROGRESS EXERCISES
Check Your Progress 1
1) A non-linear difference equation is an equation that relates the value of a
variable in one period to the value(s) of the same variable in past
periods, where the shape of the function depicting the equation is not
limited to certain specific shape. A linear difference equation is a special
case of a non-linear difference equation.
2) A phase diagram plots the values of a variable on the vertical axis and
the values of the same variable in the previous period on the horizontal
axis. It is used to depict first-order difference equations. There is a 450
line, and the intersection of the function with the 450 line helps us know
about equilibrium points.
3) (a) A linear difference equation has one equilibrium point; a non-linear
equation has multiple equilibria. (b) an attractor is a stable equilibrium
whereas a repellor is an unstable equilibrium. 311
Difference Equations Check Your Progress 2
1) We linearise a non-linear difference equation to be able to make
statements about the stability of equations.
2) We can use first-order Taylor expansion to linearise a non-linear
difference equation around its equilibrium value
Check Your Progress 3
1) The fundamental non-linear difference equation in the Solow model
relates the capital-labour ratio to its value in the previous period. This is
the basic function whose phase diagram is drawn.
2) (a) The basic shape of a chaotic dynamic system is quadratic with a hill
shaped curve, though not all quadratic functions have chaotic behavior.
(b) A limit cycle shows a periodic regular cyclical behavior with the
interval within which the function can take its values being limited.
Chaos is important in Economics because it shows how deterministic equations
can give rise to irregular aperiodic cyclical behavior. Chaos is different from a
random process in that a random process’ future values cannot be predicted
(that is the nature of the variable being random!) while a chaotic dynamic
process’ future values can be predicted.

312
 Glossary
GLOSSARY
Adjoint of a matrix: The adjoint of a square matrix A is defined as the
transpose of the matrix of the cofactors of the determinant of A.
Anti derivation: Reverse of derivation
Anti derivative: Integral
Cartesian coordinates: A system in which points on a plane are identified by
an ordered pair of numbers, representing the distances to two or three
perpendicular axes.
Continuity: A function is continuous if its graph can be drawn on a p without
lifting the pen or the pencil with which it is being drawn.
Constant of Integration: Arbitrary constant C which gets added to F(x)
Constraint: it is a side-condition that has to be fulfilled in some optimisation
exercises. It changes the value of the global optimum that would have prevailed
in the absence of such a constraint.
Cofactor: The cofactor of any element aij of the determinant of A denoted by
Cij, is in fact a signed (+ or -) minor. The sign of the minor is determined by the
rule: Cij = (-1)i + jMij.
Constant: A quantity that remains fixed in the context of a given problem or
situation.
Cobweb Model: A model of quantity demanded and supplied where demand
depends on current price while quantity supplied depends on price prevailing
one period earlier.
Calculus: A branch of mathematics which deals with change. It was described
as the mathematics of change, motion and growth.
Coordinate plane: The plane determined by a horizontal number line, called
the x-axis, and a vertical number line, called the y-axis, intersecting at a point
called the origin. Each point in the coordinate plane can be specified by an
ordered pair of numbers
Closed input-output model: In such a model, the household sector is treated
as one of the industries and thus the entire output of each producing sector is
absorbed by other producing sectors as secondary inputs or intermediate
products. Here, no portion of the output is sold in the market as final product.
Constrained optimisation is best possible position under some conditions
called constraints. For example a consumer maximizes his utility/ satisfaction
subject to his money income called income or budget constraint.
Continuity: A function f(x) is continuous provided its graph is continuous,
i.e., a continuous function does not have any break at any point of its graph.
Compounding: the process of accumulating the time-value of money or a
financial asset forward in time
Continuity of a Function: Function f is continuous at some point c when
the following two requirements are satisfied:
• f ( c ) must be defined (i.e., c must be an element of the domain of f ).

313
Glossary • The limit of f ( x ) as x approaches c must exist and be equal to f(c). (If
the point c in the domain of f is not a limit point of the domain, then
this condition is vacuously true, since x cannot approach c .
• Critical Point: It is also called stationary point where the first
derivative is zero. It can be a point of:
a) Maxima when slope changes from 0 to less than zero.
b) Minima when slope changes from 0 to more than zero.
c) Where the curvature of the curve changes either from converse to
concave or from concave to converse. This point is called the point
of inflexion.

Definite Integral: Here the numerical value of integral is a definite number,

arbitrary constant is eliminated
Dependent Variable: in an equation or function, the variable whose value
depends on that of other variable(s)
Discounting: calculating the present value of a future amount.
Dynamic Analysis: a type of analysis in which the objective is to either trace
out the time paths of the variables, or to determine whether, given sufficient
time, these variables will tend to converge to some equilibrium values.
Determinant: The determinant is a number (scalar) defined for a square matrix
only. It is conventionally represented by enclosing the elements of the
corresponding matrix within two vertical straight lines.
Discontinuity: If there is a jump or jerk in the graph at a point a over a range,
the function is discontinuous
For finding the derivative of a function, the function must be continuous.
Diagonal matrix: A diagonal matrix is a square matrix, where there is at least
one non-zero element on the principal diagonal and all the off-diagonal
elements are zeros.
Dimension or order of a matrix: The number of rows and columns of a
matrix constitutes its dimension or order.

Domain of a Function: In a function f ( x ) = y , the set X of input values is

called the domain of f , and the set Y where f takes output values is called
the codomain.
Differential Calculus is a branch of calculus which tries to study the
phenomenon of finding a derived function – a function derived from a given
function.
Derivative expresses the idea of change and is derived function on the basis of
a condition that the function is continuous. It is also called differential
dy
coefficient if the function is y = f ( x ) .
dx
Everywhere Continuous Function: Continuous at every point of its domain.
Endogenous Variable: in a model, the variables, whole values have to be
determined as solutions to the model
314
Element of a matrix: Each member of a matrix is called an element of the matrix. Glossary

Endogenous variable: It is a variable whose value is given from within a

given model.
Exogenous variable: It is a variable whose value is determined from outside a
given model.
Hawkins-Simon Condition: This is the condition for positive levels of output,
when aij’s are expressed in physical units. The condition requires that all
principal minors of the technology matrix must be positive.
Homogeneous System of Equations: A system of linear equations is
homogeneous if each constant to the right of the equality equals zero.
Homogeneous difference equation: A difference equation is homogeneous if
its constant term (the term containing no y) is zero, otherwise it is non-
homogenous.
Function : It is the rule of correspondence between dependent variable and
independent variable (s) so that for every assigned value to the independent
variable, the corresponding unique value for the dependent variable is
determined.
Indefinite Integral: Integral involving an arbitrary constant. It cannot have a
definite numerical value
Integration by substitution: When you make substitution for given
substitution integrand and use function of a function rule to evaluate integral it
is called integration by substitution
Integration-by-parts: a technique of working out integral of a product of two
function
Input coefficient matrix: It is a matrix of various secondary inputs required
by different producing sectors per unit of their output.
Input-output model or input-output transactions matrix: It is a formulation
that focuses on the interdependence of the producing sectors and, the
interaction between the producing sectors on one hand and household sector of
an economy on the other hand.
Intersection of sets : The intersection, written as A B, of two sets A and B is
the set of all elements that belong to both A and B.
Inconsistent Equations: A system of equations that yields no solution
Independent Variable: in a function or model, the variable which is/are a
causal variable(s), which influence the dependent variable.
Identity Matrix: It is a square matrix with 1s in the principal diagonal and
zeros in the off-diagonal places.
Inverse of a matrix: Given a square matrix A, if there exists another square
matrix B, such that, AB = BA = I, where I is an identity matrix, then B is said
to be the inverse of the matrix A. For a matrix A, its inverse is generally
denoted by A-1.
Integral Calculus is the other branch of calculus which finds the original
function whose derivative is given. It is opposite of differential calculus and is
called anti-derivative also.
315
Glossary Linearly Dependent Equations: These are a type of equations in a system of
equations that may be derived from each mother by a series of linear operations
Limit of a Function at a Point: Suppose f is a complex-valued function, then
we write lim f ( x ) = L if and only if for every ε > 0 there exists a δ > 0 such
x→ p

that for all real numbers x with 0 < x − p < δ , we have f ( x ) − L < ε .

Limit of a Sequence: provides a rigorous definition of the idea of a sequence

converging towards a point called the limit. Suppose you have a sequence of
points (i.e. an infinite set of points labelled using the natural numbers) in some
sort of mathematical object (for example the real numbers) which has a concept
of nearness (such as "all points within a given distance of a fixed point"). A
point L is the limit of the sequence if for any prescribed nearness, all but a
finite number of points in the sequence are that near to L.
Lagrangian Function: A function constructed to solve constrained
optimization problems by combining the objective function and the constraint
Lagrange multiplier : A quantity in constrained optimization used in the
Lagrangian function; it measure the shadow price of the variable in the
objective function.
Limit: A concept used to describe the behavior of a function as its argument
(independent variable) either "gets close" to some point, or as it becomes
arbitrarily large; or the behavior of a sequence’s elements as their index
increases indefinitely. Suppose f ( x ) is a real function and c is a real number.
Then we write lim f ( x ) = L to mean that f ( x ) can be made to be as close to
x→c

L as desired by making x sufficiently close to c . In that case, we say that "the

limit of f of x , as x approaches c , is L".
Limit of a function: The idea if limit is closely connected with the idea of
approximation and limit of a sequence. It is value of one variable approaching
or approximating as another variable approaches or approximates a specific
value. It can be a finite or infinite value. It a function approaches a finite
number. The limit is finite; if it approaches +∞ or − ∞ it is the case of infinite
limit.
L' Hospital Rule: It is a rule for finding limits of function/s which lead to
0 ∞
undefined/ indeterminate cases like ∞ − ∞, , etc. It states that if the original
0 ∞
function is not differentiable, we can try finding limits of its derivative/s.
f ( x) f ′ ( x) f ′′ ( x)
Lt = Lt = Lt = = ......
x→ a g ( x) x→ a g ′ ( x) x→ a g ′′ ( x)
Market model: It is a formulation that seeks to explain the determination of
equilibrium prices and quantities of different but related commodities.
Model: It is a set of equations, functional relationships and identities that seeks
to explain some real world phenomenon.
Matrix: It is a rectangular array of numbers.
Matrix operations: The basic operations of addition and multiplication on
matrices are called matrix operations.

316
Minor: The minor of any element aij of the determinant of some matrix A, Glossary
denoted by Mij, is the sub-determinant obtained by deleting ith row and jth
column of the determinant of A.
National income model: It is a formulation that seeks to present and explain
the relationships among national income and related aggregates and their
determination.
Non-singular matrix: A square matrix A is said to be non-singular, if its
determinant is not zero.
Null matrix: It is a matrix whose elements are all zero. It need not be a square
matrix.
Overdetermined System of Equations: A system of m independent linear
equations which is not inconsistent with n unknowns is overdetermined if m > n
Open input-output model: In such a model, the producing sectors interact
with household sector of an economy through their purchase of primary inputs
and sales of final products.
Optimisation: It is the best possible position under the given circumstances. It
can be a point of either maximum/ maxima or Minimum/ minima e.g. obtaining
maximum output with the minimum factor cost. A student wishes to maximise
marks with minimum efforts in terms of study hours.
Order of a difference equation: The order of a difference equation is the
order of the higher difference it contains
One-sided limit : Either of the two limits of a function f ( x ) of a real variable
x as x approaches a specified point either from below or from above. We write
either lim f ( x ) or lim f ( x ) for the limit as x approaches a from above (or
x→ a+ x↓ a

"from the right"), and similarly, lim− f ( x ) or lim f ( x ) for the limit as x
x→ a x↑ a

approaches a from below (or "from the left").

Parameter: A quantity that retains the same value throughout any particular
problem but may assume different values in different problem.
Partial derivative: Ordinary derivatives apply to functions of one variable:
y = f ( x ) . Partial derivative refer to finding differential coefficient of functions
of two or more variables like Z = f ( x, y ) or P = f ( L , K ) . Here we
differentiate dependent variable (say z ) with respect to either x or y by
keeping/ assuming the other variable constant.
Primary inputs or factors of production: They are the basic participants of a
production process. For their contribution to production, the net value of
production is distributed among them as factor payments.
Quadratic Form: a form is a polynomial expression in which each component
term has a uniform degree. In a quadratic form, each term is of the second
degree.
Reduced Form: in a model depicted by a set of simultaneous equations, where
as a solution to the system, the endogenous variables are expressed as functions
only of the parameters
Range of a Function: Set of all output values of a function f ( x ) = y . The
range of f is a subset of the codomain Y.

317
Glossary Relative Maxima: the set of points in whose neighbourhood only the function
is maximised or minimised.
Right and Left Hand Limits: Take the limit of f as x approaches a is L .
Note that x may approach a from above (right) or below (left), in which case
the limits may be written as lim+ f ( x ) = L or lim− f ( x ) = L respectively. If
x→a x→a
both of these limits are equal to L then this can be referred to as the limit of
f ( x ) at a . Conversely, if they are not both equal to L then the limit does not
exist.
Set : A collection of objects or elements viewed as a single entity is called a
set.
Scalar: An ordinary number is called a scalar. It is a matrix of d
Singular matrix: A square matrix A is said to be singular, if its determinant is
zero.
Square matrix: A matrix, in which the number of rows and columns are equal,
is called a square matrix.
Slope of a function: the ratio of the change in the dependent variable to a
change in the independent variable.
Stationary Value: Point at which optimum is found
Sequence : An ordered list of objects (or events). For example, (X, Y, Z) is a
sequence of letters that differs from (Y, Z, X), as the ordering matters.
Series: A series is a special type of sequence. For example, a series may be
obtained by summing the first n terms of a sequence.
Symmetric matrix: It is a square matrix whose transpose is the matrix itself.
Secondary inputs or intermediate products: These are the products of other
producing unit used by a given producing sector for its own production.
Transpose of a matrix: The transpose of a matrix, say A, is defined as a
matrix that is obtained by interchanging the rows and columns of A.
Technology matrix: This is a matrix, that is obtained by subtracting a given
input coefficient matrix from an identity matrix of appropriate dimension. This
matrix is supposed to reflect the technology.
Union of sets : The union written as A B, of two sets A and B is the set
whose elements belong to A or to B or to both.
Vector: A vector is a matrix consisting of only one row or one column.
Young's Theorem: States that the order of derivative does not matter that is
f xy = f yx .

318
 Glossary
SUGGESTED READINGS
Sydsaeter, Knut and Hammond, Peter J. (1995) Mathematics for Economic Analysis,
Pearson Education, Inc, NOIDA, India.

Hoy, Michael, Livernois, John, McCenna, Chris, Rees, Ray, and Stengos, Thanasis
(2001) Mathematics for Economics the MIT Press, Cambridge, Massachusetts, USA.
Indian Reprint (Prentice-Hall of India Pvt Ltd, New Delhi India

Chiang, Alpha C. and Wainwright, Kevin (2005) Fundamental Methods of

Mathematical Economics 4th edition, McGraw-Hill International edition, Boston,
USA.

Renshaw, Geoff (2009) Maths for Economics, Oxford University Press, New York,
USA. Indian Reprint available

319