This action might not be possible to undo. Are you sure you want to continue?
)
GUOQIANG TIAN
I.
THE NATURE OF MATHEMATICL
ECONOMICS
Mathematical economics is an approach to economic analysis, in which the economists make use of mathematical symbols in the statement of the problem and also draw upon known mathematical theorems to aid in reasoning. The purpose of this course is to introduce the most fundamental aspects of the mathematical methods such as those matrix algebra, mathematical analysis, and optimization theory. 1.1 Mathematical Versus Nonmathematical Economics
Since mathematical economics is merely an approach to economic analysis, it should not and does not differ from the nonmathematical approach to economic analysis in any fundamental way. The difference between these two approaches is that in the former, the assumptions and conclusions are stated in mathematical symbols rather than words and in equations rather than sentences. Advanta2es of the Mathematical Approaches
(1)
(2) (3)
the analysis is more rigorous; it allows us to treat the general nvariable case; and the "language" used is more concise and precise.
III. EQUILIBRIUM
3.1 The Meaning
ANALYSIS IN ECONOMICS
of Equilibrium One definition
Like any economic term, equilibrium can be defined in various ways. here is that an equilibrium
is a constellation of selected interrelated variables so adjusted to one
another that inherent tendency to change prevails in the model which they constitute. In essence, an equilibrium for a specific model is a situation that is characterized lack of tendency to change. as statics. The fact that an equilibrium by a
It is for this reason that the analysis of equilibrium is referred to implies no tendency to change may tempt one to
conclude that an equilibrium necessarily constitutes a desirable or ideal state of affairs. This chapter provides two examples of equilibrium. a market under given demand and supply conditions. One is the equilibrium attained by
The other is the equilibrium of national
income under given conditions of consumption and investment patterns. 3.2 Partial Market Equilibrium  A Linear Model
In a staticequilibrium
model, the standard problem is that of finding the set of values conditions of the model.
of the endogenous variables which will satisfy the equilibrium PartialEquilibrium Three variables
Market Modela model of price determination in an isolated market.
Qd
P
= =
the quantity demanded of the commodity; the price of the commodity. Condition:
Q. = the quantity supplied of the commodity;
The Equilibrium The model is
Qd
=
Q •.
Qd = Q. Qd = a  bp Q. = c + dp
> 0) (c, d > 0)
(a, b
p
The slope of Qd
=
b, the vertical intercept
=
a.
The slope of Q. = d, the vertical intercept = c. Note that, contrary to the usual practice, quantity rather than price has been plotted vertically in the Figure. One way of finding the equilibrium equations through substitution. From Q. is by successive elimination of variables and
=
Qd, we have
a  bp and thus
=
c
+
dp
(b + d)p = a + c. Since b + d ~ 0, then the equilibrium price is

a+c
p  b+d
The equilibrium quantity can be obtained by substituting p into either Q. or Qd:

32
Q_
adbe b+d
Since the denominator
(b+d) is positive, the positivity of Q requires that the numerator (adbe)

> O. Thus, to be economically meanful, the model should contain the additional restriction that ad > bc.
3.3 Partial Market Equilibrium  A Nonlinear Model Suppose a model is given by
The partial market model can be nonlinear.
Qd = Qd Qd = 4  p2
Q.
=
4p  1 stated, this system of three equations can be reduced to a single equation by
As previously substitution. 4  p2 or
=
4p  1
p2 + 4p  5
=
0 In general, given a quadratic equation in the form
(a ~ 0),
which is a quadratic equation. ax2 + bx +
C
=
0
its two roots can be obtained from the quadratic formula:
b ± Jb2 4ae 2a
where the "+" part of the
"±" sign yields x, and the "" part yields X2·


33
Thus, by applying the quadratic formulas to p2 = 5, but only the first is economically The Graphical Solution admissible,
+ 4p  5 = 0, we have
PI =
1 and P2
as negative prices are ruled out.
/
/2
/
/
/
/
/
/
/
:
,..
1
3.4
General Market Equilibrium
In the above, we have discussed methods of an isolated market, wherein the Qd and Q.
of a commodity are functions of the price of that commodity alone. would normally exist many substitutes and complementary only of the price of the commodity the model. equilibrium goods.
In the real world, there
Thus a more realistic model As a into
for the demand and supply functions of a commodity should take into account the effects not itself but also of the prices of other commodities. must enter endogenously are simultaneously commodities result, the price and quantity variables of multiple commodities Thus, when several interdependent would require the absence of excess demand,
considered, between the
which is the difference
demand and supply, for each and every commodity included in the model. equilibrium condition of an ncommodity commodity, in the form (i= 1,2, ... ,n) 34
Consequently,
market model will involve n equations, one for each
where Qdi
=
Qdi (PI,P2,···,Pj
and Q.i
=
Q.i (P1,·.. ,Pj are the demand and supply functions of
commodity i, and (P1,P2, ... ,Pj
are prices of commodities.
Thus solving n equations for P:
we obtain the n equilibrium prices Piif a solution does indeed exist. derived from the demand or supply functions. TwoCommodity Market Model let us consider a twocommodity

And then the Qi may be
To illustrate the problem, demand and supply functions. Qdl  Q&l = 0 Qdl Q.l
market model with linear
In parametric
terms, such a model can be written as
= =
<10
+ a.P, + a2P2
bo + b.P, + b2P2
Qd2  Q.2 Qd2 = ao Q.2
=
0
=
+ alPI + a2P2 f30 + f3IPI + f32Pz
By substituting the second and third equations into the first and the fifth and sixth equations into the fourth, the model is reduced to two equations in two variables:
If we let
ci
=
a,  b,
(i=O,1,2),
35
the above two linear equations can be written as
which can be solved by further elimination of variables. The solutions are
For these two values to make sense, certain restrictions should be imposed on the model. First, we require the common denominator
C!"'2 
c21', ;r:. O.
Second,
to assure positivity,
the
numerator must have the same sign as the denominator. Numerical Example Suppose that the demand and supply functions are numerically Qd' Q., Qd2 Q.2 as follows:
= = = =
10  2P, 2 1
+ P2
+ 3P, 15 + P,  P2
+
2P:,
By substitution, 5P,  P2
PI
we have
=
12
+ 3P2
=
16 The solutions for the equilibrium prices and quantities are 6417, and Q2

which are two linear equations.

P,
=
52/14, P2

=
92/14, Q,

=
=
36
8517.
Similarly, for the ncommodity
market model, when demand and supply functions are In the above, we assume that an equal number However, some very simple examples should
linear in prices, we can have n linear equations. of equations and unknowns has a unique solution. existence of a unique solution. For the two linear equations,
convince us that an equal number of equations and unknowns does not necessarily guarantee the
j
x+y x+y
8 9
we can easily see there is no solution. The second example shows a system has an infinite number of solutions: 2x + Y = 12 4x + 2y
= 24
one equation is redundant and may be dropped from the system. Any
These two equations are functionally dependent, which means that one can be derived from the other. Consequently, pair (x,y) is the solution as long as (x,y) satisfies y
=
12  x. In general, there is no independent
Now consider the case of more equations than unknowns. solution. equations, the solution exists and is unique. 2x + 3y
y
But, when the number of unknowns equals the number of functional
The following example shows this fact.
=
58 18
=
x + y = 20 Thus for simultaneousequation a unique (or determinate) model, we need systematic methods of testing the existence of These are our tasks in the following chapters.
solution.
37
3.5
Equilibrium
in NationalIncome
Analysis As a simple
The equilibrium analysis can be also applied to other areas of economics. example, we may cite the familiar Keynesian nationalincome model,
i=
C
C+
10 + Go
(equilibrium
condition) function) national income and consumption
=
a+
bY
(the consumption for the endogenous variables
where
"I
and C stand respectively,
expenditure,
and 10 and Go represent the exogenously determined investment and
government expenditures. Solving these two linear equations, consumption expenditure: we obtain the equilibrium national income and
y

a + 10 + Go
Ib
38
IV.
LINEAR
MODELS
AND MATRIX
ALGEBRA

P and Q are relatively simple, even though a number of parameters are involved. are incorporated cumbersome and unwieldy.

From the last chapter we have seen that for the onecommodity
model, the solutions for As more and
more commodities of simultaneous
into the model, such solution formulas quickly become
We need to have new methods suitable for handling a large system Such a method is found in matrix algebra. (1) It provides a compact way of (2) It leads to a way of testing the concept closely related to that of a
equations.
Matrix algebra can enable us to do many things. writing an equation system, even an extremely large one. existence of a solution by evaluation of a determinantsa matrix. 4.1
(3) It gives a method of finding that solution if it exists. Matrix and Vectors In general, a system of m linear equations in n variables
(X1,X2, ...
.x.) can be arranged
into such a formula. allx1
~IXI
+ a12x2 +
+ a22x2 +
+ a1nxn + ~Xn
= =
d, d, (4.1)
where the doubledsubscripted and attached to the jth variable Example:
symbol aij represents the coefficient appearing in the ith equation
Xj
and d, represents the constant term in the jth equation. linear market model can be written+after eliminating the
The twocommodity
quantity variablesvas a system of two linear equations.
C1P1
'YIPI
+
+
~P2
'Y2P2
=
co
= 'Yo
Matrix as Arrays There are essentially three types of ingredients in the equation system (4.1). is the set of coefficients aij; the second is the set of variables of constant terms d., ... ,dm• them, respectively, as, A, x, and d, then we have
Xl'X2' ... 'Xn;
The first
and the last is the set arrays and label
If we arrange the three sets as three rectangular
all
an
... ...
aln a2n
Xl
dl d
A

~l
~
am2
X

~
X

d2 dm
(4.2)
a
ml
...
a
mn
n
Example:
Given the linearequation
X3
system
6x1 + 3x2 +
=
22 12
x, +
4X2  2X3
=
we can write
A  [~ : ~2]
4 1 5
X

[::] X3
d  [~~]
10
Each of three arrays in (4.2) constitutes a matrix. A matrix is defined as a rectangular array of numbers, parameters, shorthand device, the array in matrix A can be written more simply as or variables. As a
.i 1,2, ,n [ J 1,2, ,m
1
Vectors as Special Matrices The number of rows and the number of columns dimension of the matrix. where m=n, the matrix is called a square matrix. in a matrix together define the
For instance, A is said to be of dimension m x n. In the special case
42
If a matrix contains only one column (row), it is called a column (row) vector. , For notation purposes, primed symbol. a row vector is often distinguished from a column vector by the use of a
Remark.
A vector is merely an ordered ntriple and as such it may be interpreted as a point in
an ndimensional space. With the matrices defined in (4.2), we can express the equation system (4.1) simply as Ax However,
=
d. How do we multiply two
the equation Ax = d prompts at least two questions.
matrices A and x? What is meant by the equality of Ax and d? Since matrices involve whole blocks of numbers, the familiar algebraic operations defined for single numbers are not divertly applicable, and there is need for new set of operation rules. 4.2 Matrix Operations
The Equality of Two Matrices. A = B if and only if a, = bij for all i=1,2, ... ,n,j=I,2, Addition and Subtraction of Matrices. ... ,m.
i.e. the addition of A and B is defined as the addition of each pair of corresponding Remark. Example:
elements.
Two matrices can be added (equal) if and only if they have the same dimension.
43
Example:
The Subtraction of Matrices A  B is defined by
Example:
[ 19 3] _ [6 8] _ [13 5]
2 0 13 1 3
Scalar Multiplication
i.e. to multiply a matrix by a number is to multiply every element of that matrix by the given scalar. Example:
Example:
44
Multiplication
of Matrices condition for multiplication AB
Given two matrices Arnxn and Bpxq, the conformability product AB will be defined if and only if n = p. dimension m x q, The product AB is defined by AB = C
is that the column dimension of A must be equal to be the row dimension of B, i.e., the matrix If defined, the product AB will have the
with
c
ij

ailblj
+
ai2b2j
+ ... +
aEbnj

L
II
n
aifblj
Example:
Example:
B2t2

1 0 [47
1
AB _ [ 3 + 20 35]
4+24 42
=[
17 35] 20 42
45
Example:
u'v 
U,V,
+llzv2+···+unvn

L
iI
n
u.v,
I
I
This can also be described by using the concept of the inner product of two vectors u and v.
Example:
all
a'2 an
...
a'll
a:>n
x, x
A

a2,

x,
a ml
a m2
... a mn
Xn
Example:
Given
u  [ ~]
and
Vi
=
[1
4 5]
uv 
3 x 1 3x4 [ 2xl 2x8
15 ]
10
It is important to distinguish the meanings of uv' (a matrix larger than 1 x 1) and u'v (a 1 x 1 matrix, or a scalar).
46
4.3
Linear Dependence of Vectors A set of vectors vl, Vnis said to be linearly dependent if and only if anyone of them
... ,
can be expressed as a linear combination of the remaining vectors; otherwise they are linearly independent. Example: The three vectors
are linearly dependent since V3is a linear combination of v! and v2:
where
0  [ ~]
represents a zero vector.
An equivalence
definition of linear dependence
is: a set of mvectors
V!'V2'.. ·'Vn is
linearly dependent if and only if there exists a set of scalars khk2,···,k" (not all zero) such that
If this equation can be satisfied only when k, = 0 for all i, these vectors independent. 4.4 Commutative, Associative, and Distributive Laws
are linearly
The commutative and associative laws of matrix can be stated as follows: Commutative Law A+B=B+A 47
Associative Law Matrix Multiplication Matrix multiplication AB ;:e BA
(A
+ B) + C
=
A
+ (B + C)
is not commutative,
that is
Even when AB is defined, BA may not be; but even if both products are defined, AB
=
BA may not still hold.
Example:
Let
A
[:
~]
then
AB 
[ 12 13]
24 25
,
but
BA 
[ 3 4]
27 40
.
The scalar multiplication leA = Ak if k is a scalar. Associative Law
of a matrix does obey.
The commutative law:
(AB)C = A(BC)
if A is rnxn, B is nxp, and C is pxq. Distribution Law A(B (B
+ C)
= AB
+ C)A
=
+ AC BA + CA
[premultiplication [postmultiplication
by A] by A]
48
4.5
Identity Identity
Matrices Matrix
and Null Matrices is a square matrix with ones in its principal diagonal and zeros
everywhere else.
It is denoted by I or In, in which n indicates its dimension.
Fact 1 Given an rnxn matrix A, we have
Fact 2 (AI)B
=
AB
Null Matrices.
A null+or zero matrix=denoted
by 0, plays the role of the number O. A null Unlike I, the zero matrix is not
matrix is simply a matrix whose elements are all zero. restricted to being square. Amxn + Omxn = Amxn Amxn Onxp Oqxm Amxn Remark. CD
Null matrices obey the following rules of operation.
= Omxp = Oqxn
CE does not imply D
=
=
E. For instance,
C  [2 3]
69
D_
[1 1]
12
E_[
2 1]
3 2
we can see that
CD  CE _
[5 8]
15 24
even though D ,,: E.
49
4.6
Transposes and Inverses The transpose of a matrix A is a matrix which is obtained by interchanging the rows and
columns of the matrix A. It is denoted by A' or AT.
Example:
For
A
[ 3 8 9]
10 4
Thus, by definition, Example:
if a matrix A is rnxn, then its transpose A' must be nxm.
Its transpose
D'
[~1073~41
=D
A matrix A is said to be symmetric if A'
Properties of Transposes a)
b)
=
A.
(A')'
(A
=A B)' = A'
+
+
B'
c)
(AB)' = B' A'
410
The property C states that the transpose of a product is the product of the transposes in reverse order. Inverses and Their Properties For a given matrix A, A' is always derivable. mayor may not exist. On the other hand, its inverse matrix
The inverse of A, denoted by AI, is defined only if A is a square
matrix, in which case the inverse is the matrix that satisfies the condition.
Remarks. 1. Not every square matrix has an inversesquareness condition for the existence of an inverse. to be nonsingular, 2. 3. 4. If A is nonsingular, is a necessary but not sufficient If a square matrix A has an inverse, A is said
if A possesses no inverse, it is said to be a singular matrix. then A and AI are inverses of each other, i.e., (AI)"I
=
A.
If A is nxn, then A:' is also nxn. The inverse of A is unique. Proof. B Let Band C both be inverses of A. Then BAC
=
BI
=
= =
1.
IC
=
C.
5.
AA·I Proof.
=
I implies AlA
We need to show that if AAl
I, then B
=
=
I, and if there is a matrix B such that BA
Al.
Postmultiplying
both sides of BA
=
I by AI, we have BAAl
=
=
Al
and thus B = AI.
Example:
let
A
[0321]
411
So B is the inverse of A. nxn.
Suppose A and B ar nonsingular
matrices with dimension
6.
(a) (b)
(AByl (A')I
= B·Ikl = (A:'):
System
Inverse Matrix and Solution of LinearEquation The application
of the concept of inverse matrix to the solution of a simultaneousConsider
equation is immediate and direct. Ax
=
d matrix, then premultiplying both sides of Ax = d, we have
If A is a nonsingular AIAx
=
kid Methods
So, x = kid is the solution of Ax = d and the solution is unique since kl is unique.
of testing the existence of the inverse and of its calculation will be discussed in the next chapter.
412
V. LINEAR MODELS AND MATRIX ALGEBRA (CONTINUED)
In Chapter 4, it was shown that a linearequation notation. Furthermore, system can be written in a compact
such an equation system can be solved by finding the inverse of the
coefficient matrix, provided the inverse exists. This chapter studies how to test for the existence of the inverse and how to find that inverse.
5.1
Conditions for Nonsingularity of a Matrix
As was pointed out earlier, the squareness condition is necessary but not sufficient for
the existence of the inverse Al of a matrix A. Conditions for Nonsingularity When the squareness condition is already met, a sufficient condition for the
nonsingularity of a matrix is that its rows (or equivalently, squareness and linear independence conditions.
its columns) are linearly independent. are that the matrix satisfies the
In fact, the necessary and sufficient conditions for nonsingularity
An nxn coefficient matrix A can be considered an ordered set of row vectors:
all
al2 a 22
...
...
a
ln
I VI I V2
A

a
21
a2n
anI
«.
...
a
M
v,
I
where
v:
= [a;I,~, ... ,aJ, i= 1,2, ... ,n. For the rows to be linearly independent,
for any set of
scalars k.,
L k1v1
n j .. l
0
if and only if k, = 0 for all i.
Example:
For a given matrix,
A[~~~1,
6 8 10
since v;
=
2v~
+
Ov], so the matrix is singular.
Rank of a Matrix Even though the concept of row independence has been discussed only with regard to square matrices, it is equally applicable to any rnxn rectangular matrix. If the maximum number of linearly independent rows that can be found in such a matrix is 'Y, the matrix is said to be of rank 'Y. The rank also tells us the maximum number of linearly independent columns in the said matrix. The rank of an rnxn matrix can be at most m or n, whichever is smaller. an n xn nonsingular matrix A has n linearly independent rows (or
By definition, columns); consequently be nonsingular. 5.2
it must be of rank n. Conversely,
an n x n matrix having rank n must
Test of Nonsingularity
by Use of Determinant we can make use of the concept
To determine whether a square matrix is nonsingular, of determinant. Determinants and Nonsingularity of a square matrix A, denoted by Determinants
The determinant matrix:
I A I , is a uniquely
defined scalar For a 2 x2
associated with that matrix.
are defined only for square matrices.
A
_
all
aI2],
[ a22 a22
its determinant
is defined as follows:
IAI 
In view of the dimension of matrix A, determinant.
IA I
as defined in the above is called a secondorder
52
Example:
Given
A
5 [ 10 4]
8
and
B_
[3o 5 ],
1
then
IA I 
1
10 8
41 5
 50  32  18
IBI
_13o
5 1
1 3 
5xO 
3
Example:
A_
[2
8 24
6]
Then its determinant
IA I
_12 8
6 24
1
2x24  6x8  48  48  0
This example shows that the determinant will be seen, the value of a determinant calculation of the inverse AI, if it exists. Evaluating a ThirdOrder Determinant
is equal to zero if its rows are linearly dependent.
IA I
As
can serve not only as a criterion for testing the of matrix A but also as an input in the
linear independence of the rows (hence nonsingularity)
For a 3 x 3 matrix A, its determinants
have the value
all
a12 a13
 all
IAI 
a21 a22 a23 a31 an a33
a22 a23 a32 a33
 al2
Ozl a23
a31 a33
+
a13
Ozl a22
a31 a32
53
Example:
213
456
789
=
determinant,
90
+ 42 + 96  105  36  96
= 9
This method of crossdiagonal but unfortunately
multiplication provides a handy way of evaluating a thirdorder it is not applicable to determinants of orders higher than 3. For
the latter, we must resort to the socalled "Laplace expansion" of the determinant. Evaluating an nthOrder Determinant by Laplace Expansion
The minor of the element aij of a determinant determinant, and a13 are it has a value.
I A I , denoted
by
deleting the ith row and jth column of the determinant
I A I.
I Mij I
can be obtained by minors of all, a12,
For example, for a 3x3 determinant
I A I , the
Since a minor is itself a
A concept closely related to the minor is that of the cofactor. is a minor with a prescribed
A cofactor, denoted by it is defined by
I Cij I ,
algebraic sign attached to it. Formerly,
54
Thus, if the sum of the two subscripts i and j in 1 ~ odd, then determinant
1 Cij 1
as
=
I
Mij
I·
1 is even, then 1 c, 1
=
1 Mij I.
If it is
Using these new concepts,
we can express
a thirdorder
1A 1 =
all 1 Mll
= all
1 Cll
1  al21M21 + a1311l1b 1 1 + al21 cI21 + al31 cI31
determinant serves to reduce the evaluation In general, the Laplace of the process will determinants.
The Laplace expansion of a thirdorder
problem to one of evaluating only certain secondorder each of which is of the (nl )st order, secondorder Formerly, determinants.
expansion of an nthorder determinant will reduce the problem to one of evaluating n cofactors, and the repeated application methodically lead to lower and lower orders of determinants, the value of a determinant eventually culminating in the basic
Then the value of the original determinant can be easily calculated.
1 A 1 of order n can be found by the Laplace expansion
of any row or any column as follows:
IAI
L
n
aijlCifl
[expansion by the ith row]
jI
L
n
iI
aij I Cifl
[expansion by the jth column].
Even though one can expand calculation is concerned,
1 A 1 by any row or any column,
as the numerical
a row or column with largest number of Os or Is is always preferable
for this purpose, because a 0 times its cofactor is simply O.
561 Example: For the
IA I 
2
3 0,
7 3 0
the easiest way to expand the determinant is by the third column, which consists of the elements 1, 0, and O. Thus,
55
IAI 
1(1)· 31 1
2
3 1  6  21  27 7 3
5.3
Basic Properties of Determinants
Property 1. The determinant of a matrix A has the same value as that of its transpose A', i.e.,
IAI = IA'I·
Example:
IA I
I: : 1I: : 1
ad  be
lA' I
ad  eb 
IA I
Property II. The interchange of any two rows (or any two columns) will alter the sign, but not the numerical value of the determinant.
Example:
I::1I: ~I
ad  be,
but the interchange of the two rows yields
cb=ad
 tab=cdy.
Property III. The multiplication value of the determinant
of anyone
row (or one column) by a scalar k will change the
kfold, i.e. for
IA I ,
56
all
al2
...
all.
all
al2
... all.
/call
/cai2
...
/caill
k
ail
ai2
... aill
 klAI
ad
an2
...
a""
anI
an2 ... a""
In contrast, the factoring of a matrix requires the presence of a common divisor for all its elements, as in
all
al2
... ...
aln
ka., ka..
ka.; ka22
... ...
ka., ka2n
k
a21
aZ2
a2n
an!
an2 ...
a mn
ka m! ka m2 ... ka mn
Property IV. The addition (subtraction) of a multiple of any row (or column) to (from) another row (or column) will leave the value of the determinant unaltered.
Example:
l
ab c +ka
d+kb
1
a(d+kb)

b(c+/ca)
adbeI:
determinant will be zero.
:1
Property V. If one row (or column) is a multiple of another row (or column), the value of the
Example:
I ~ ~ 1
kab  kab  0
57
.. .__._

Remark.
Property V is a logic consequence of Property IV. The basic properties just discussed are useful in several ways. For one thing, they can multipliers may be to
be of great help in simplifying the task of evaluating determinants. reduced to much simpler and simpler numbers.
By subtracting
of one row (or column) from another, for instance, the elements of the determinant
If we can indeed apply these properties
transform some row or column into a form containing mostly Os or Is, Laplace expansion of the determinant will become a much more manageable task. Determinantal Criterion for Nonsingularity concern is primarily By Property to link the linear dependence of rows with the is
Our present vanishing
of a determinant.
I, we can easily see that row independence
equivalent to column independence. Given a linearequation system Ax
=
d, where A is an nxn coeft1cient matrix, we have
IA I ~
0 ~ A is row (or column) independent ~ A is nons ingular ~ A:' exists ~ a unique solution
x=
kId exists. a convenient criterion for testing the
Thus the value of the determinant nonsingularity
of A provides
of matrix A and the existence of a unique solution to the equation system Ax
=
d.
Rank of a Matrix Redefined The rank of a matrix A was earlier defined to be the maximum independent rows in A. of the determinant, nonvanishing determinant In view of the link between row independence that can be constructed number of linearly and the nonvanishing
we can redefine the rank of an rnxn matrix as the maximum order of a from the rows and columns of that matrix.
The rank of any matrix is a unique number. Obviously, the rank can at most be m or n for a rnxn matrix A, whichever is smaller, because a determinant expressed as follows: is defined only for a square matrix. Symbolically, this fact can be
58
'Y(A) ~ min{m,n}. The rank of an rum nonsingular matrix A must be n; in that case, we may write 'Y(A) the product of two matrix, 'Y(AB) ~ min{')'(A), 'Y(B)}. 5.4 Finding the Inverse Matrix system Ax
=
n. For
If the matrix A in the linearequation the solution of the system will be by the criterion that test. Expansion of a Determinant
I A I ~ O.
x
=
d is nonsingular,
then Al exists, and of A
= kid.
We have learned to test the nonsingularity
The next question is how we can find the inverse Al if A does pass
by Alien Cofactors
We have known that the value of a determinant Laplace expansion of any row or any column as follows:
IA I
of order n can be found by the
IAI
1:
jI
n
aijlCul
[expansion by the ith row]
1:
n
aijlCijl
[expansion by the jth column]
iI
Now what happens if we replace aij by
a,'j
for i ~ i' or by
a,j',
for j ~ j' . Then we have the
following important property of determinants. Property VI. The expansion of a determinant by alien cofactors (the cofactors of a "wrong" That is, we have
row or column) always yields a value of zero.
jI
1: I C I ail ij
n
0
(i ~ i ') [expansion by i'th row and use of cofactors of ith row] (5.10)
59
L av.ICul iI
n
0
(j ~ j')
[expansion by j'th column and use of co factors of jth column]
The reason for this outcome lies in the factor that the above formula can be considered as the results of the regular expansion by the ith row (jth column) of another determinant, differs from
IA I
which
only in its i'th row (j'th column) and its ith row (jth column) and i'th row
(j'th column) are identical. Example: For the determinant
consider another determinant
If we expand
I A"I
by the second row, then we have
I:aljIC?j1
jI
3
Matrix Inversion Property VI is of finding the inverse of a matrix. For a n X n matrix A:
510
all
al2
... ...
aln a2n
A

a21 an
«:
an2
...
a nn
since each element of A has a cofactor each element
<l;j
with its cofactor however,
I C I.
ij
I C I , we can form
ij
a matrix of co factors by replacing
Such a cofactor matrix C
= [I C I] is also
ij
n x n. For
our present purpose,
the transpose of C is of more interest.
This transpose C' is
commonly referred to as the adjoint of A and is denoted by adj A. Thus:
C' 
adj A

ICIII IC 1 ICl21 I c: 1
21 2
... ...
I <. I <. I <,
IClnl IC",I
By utilizing the formula for the Laplace expansion and Property VI, we have
AC' 
n LaljlCljl n L~jlCljl jl
j
e
l
L aljlC2jl jl
n L~·IC,·I
J jI J
n
...
L aljlCnjl
n L~jlCnjl jl
jl
n
...
L jl
n
anjlCljl
L a·IC,·1
nJ .... J jI
n
...
L
jl
e
n
anjlCnjl
IAI 0 o IAI ... o
0
0 0
IAI
511
Since
I A I ~ 0,
c' AW
then
1.
Thus, by the uniqueness of A:' of A, we know
AI
_
m
adj A
c'
JAr
Now we have found a way to invert the matrix A. inverse of a square A are: (1) find C The general procedures for finding the
I A I ; (2)
find the cofactors of all elements of A and form
adj AI.
=
[I Cij I]; (3) form C' to get adj A; and (4) determine A:' by
IA
In particular,
for a 2 x 2 matrix
A
[:
~]
we have the following formula:
AI
_ adj A
JAr
[ d b ] adcb c a
This is a very useful formula.
512
Example:
A[3120l
The inverse of A is given by
1
Example:
Find the inverse of
B _ [~
~ 2 ]. 30 7
Since
IB I =
99 ,c 0, B1 exists.
The cofactor matrix is:
1~ ~ 1 I: ~ 1 I: ~I _I ~ 1~
1 7 1 2
21
6 31
9
1 I: 1 I ~
~1
1 I: ~ 1 1 1~ ~ 1

7 5
3
8 12
~1
Then,
513
Hence,
5.5
Cramer's Rule The method of matrix inversion just discussed enables us to derive a convenient way of
solving a linearequation Derivation of the Rule
system, known as Cramer's
rule.
Given an equation system Ax
=
d, the solution can be written as
x
A ld  _1_ (adj A)d
IAI
provided A is nonsingular.
Thus,
X
lAT
1
I <. I C211 I Cnl I C221 I C1nl I C2n1
...
...
I Cnll I Cn21 I «,
d1 d2 dn
...
1AT
1
514
That is, the X; is given by

Xj 
1 lATL
n
II
djlCijl
all a l2
... dl
...
... aln
lAT
an! an2
1
a
21
a22
d2 ... a2n
... d n
. ..
a
""
==
lAT
1
IAjl
where
I Aj I
is obtained by replacing the jth column of rule.
IA I
with the constant terms dl,···,dn•
This result is the statement of Cramer's Example: 5x, + 3x2 6x, 2X2
=
=8
30
IA I
_15 6
30 8
3 2
1I
28
JAIl 
1
3 _2
 84
IA21 
5 30 68
1
I
 140
Therefore,
by Cramer's
rule, we have
515
140 _ 5.
28
Note on Homogeneous A linearequation i.e. if Ax = O. If is a "trivial solution." applicable.
 Equation System system Ax = d is said to be a homogeneousequation system if d=O,
I A I ~O,
x=O is a unique solution of Ax = 0 since x = AIO = O. This
Thus, the only way to get a nontrivial solution from the homogeneous
equation system is to have
IA I
= 0, i.e. A is singular.
In this case, Cramer's
rule is not
Of course, this does not mean that we cannot obtain solutions; it means only that In fact, it has an infinite number of solutions.
we cannot get a unique solution. Example: allxl
~IX2
+
al2x2
+ a22x2
= =
0 0 As a result, one of two equations is
If
IA I
= 0, then its rows are linearly dependent. By deleting, say, the second equation, The solutions are
redundant. variables.
we end up with one equation with two
516
Solution Outcomes for Ax = d,
d~0 ~ IAI ~0 ~0 Equations dependent IAI=o Equations inconsistent An infinite number of solutions and x ~ 0 No solution exists

d =0
The solution is unique and x
The solution is unique and x=0 There is an infinite number of solutions [Not applicable]
5.6
Application to Market and NationalIncome Models
Market Model The twocommodity
CtPt 'YtPt
model described in Chapter 3 can be written as follows:
+
+
~P2
=
C,
'Y2P2 = 'Y2'
Thus
IAI 
Thus the equilibrium is given by
517
NationalIncome Y c
Model Go (a>O, O<b< into the form 1)
= c + 10 + = a + bY
These can be rearranged Yc
bY + c
= 10 + Go =a

While we can solve Y and c by Cramer's coefficient matrix.
rule, here we solve this model by inverting the
Since
A
[1 1],
b
1
then
A'
1b
1 [bIll]
Hence
[;]
Ib
1
[:
:]
[,. : G. ]
[ b;;:+~;:a]
518
VI. COMPARATIVE STATICS AND TIlE CONCEPT OF DERIVATIVE 6.1 The Nature of Comparative Statics
Comparative statics is concerned with the comparison of different equilibrium states that are associated with different sets of values of parameters and exogeneous variables. When the value of some parameter or exogeneous variable that is associated with an' initial equilibrium changes, we can get a new equilibrium. Then the question posed in the comparativestatic analysis is: How would the new equilibrium compare with the old? It should be noted that in the comparativestatics analysis we don't concern with the process of adjustment of the variables; we merely compare the initial (prechange) equilibrium state with the final (postchange) equilibrium state. We also preclude the possibility of instability of equilibrium for we assume the equilibrium to be attainable. It should be clear that the problem under consideration is essentially one of finding a rate of change: the rate of change of the equilibrium value of an endogenous variable with respect to the change in a particular parameter or exogeneous variable. For this reason, the mathematical concept of derivative takes on preponderant significance in comparative statics. 6.2 Rate of Change and the Derivative
We want to study the rate of change of any variable y in response to a change in another variable x, where the two variables are related to each other by the function
y = f(x)
Applied in the comparative static context, the variable y will represent the equilibrium value of an endogeneous variable, and x will be some parameter. The Difference Quotient We use the symbol ~ to denote the change from one point, say Xo,to another point, say x.. Thus M = x,  Xo. When x changes from an initial value Xoto a new value (x, + M), the value of the function y=f(x) changes from f(Xo)to f(Xo + M). The change in y per unit of change in x can be represented by the difference quotient
61
l1y l1x
f(Xc! l1x)  f(Xo) + l1x
Example:
y = f(x) = 3x2  4
Then f(Xo)= 3x~  4 and thus,
l1y l1x f(Xc! l1x)  f(xO> + l1x 3(Xc! llx)2  4  (3X;  4) + l1x  6Xo
+
6xat1x + 3(l1X)2 l1x
311x
The Derivative Frequently, we are interested in the rate of change of y when Ax is very small. In particular, we want to know the rate of Ay/Ax when Ax approaches to zero. If, as Ax + 0, the limit of the difference quotient Ay/Ax exists, that limit is called the derivative of the function y=f(x), and the derivative is denoted by
dy dx
E
y'
E
f'(x)
E
lim l1y ~~ l1x
Remark. Several points should be noted about the derivative: (1) a derivative is a function. Whereas the difference quotient is a function of Xoand Ax, the derivative is a function of Xo only; and (2) since the derivative is merely a limit of the difference quotient, it must also be of necessity a measure of some rate of change. Since Ax + 0, the rate measured by the derivative is in the nature of an instantaneous rate of change. Example: Referring to the function y=3x2

4 again. Since
l1y _ 6x l1x
+
311x,
62
we have
dy _ 6x. dx
6.3
The Derivative and the Slope of a Curve
Elementary economics tells us that, given a totalcost function C=f(Q), where C is the total cost and Q is the output, the marginal cost MC is defined as MC = t1CI t1Q. It is understood that t1Q is an extremely small change. For the case of a product whose quantity is a continuous variable, t1Q will refer to an infinitesimal change. It is well known that MC can be measured by the slope of the total cost curve. But the slope of the totalcost curve is nothing but the limit of the ratio t1CI t1Q as t1Q + O. Thus the concept of the slope of a curve is merely the geometric counterpart of the concept of the derivative.
C
C=f(Q)
C2

G
II II
AQ
H
•
0
Qo
QIQ2
i~
II
Q
6.4
The Concept of Limit
In the above, we have defined the derivative of a function y=f(x) as the limit of t1y/t1x as t1x + O. We now study the concept of limit. For a given function q=q(v), the concept of limit is concerned with the question: What value does q approach as v approaches a specific value? That is, as v + N (the N can be any number, say N=O, N= + 00, 00), what happens 63
to
lim g(v).
_N
When we say v .... N, the variable v can approach the number N either from values greater than N, or from values less than N. If, as v .... N from the left side (from values less than N), q approaches a finite number L, we call L the leftside limit of q. Similarly, we call L the rightside limit of q if_v ....N from the right side. The leftside limit and rightside limit of q are denoted by
lim q and lim q,
_N· _N·
respectively. The limit of q at N is said to exist if
_N"
lim q  lim q
_N"
and is denotedby lim q.
_N
Note that L must be a finite number. If we havethe situationof lim q _N
00
(or (0), we shall consider q to possess no limit or an "infinite limit." It is important to realize that the symbol 00 is not a number, and therefore it cannot be subjected to the usual algebraic operations. Graphical Illustrations There are several possible situations regarding the limit of a function, which are shown in the following diagrams.
~ ..
11;
~'___________N
V
o
~~~v
N
I
I
I

..... ~(
I
I
M
tI
I I I
I
I
I
I
I
bt
Evaluation of a Limit Let us now illustrate the algebraic evaluation of a limit of a given function q=g(v).
Example:
Given q = 2 +
lim q "'2.
yo()
v', find
lim q.
yo()
It is clear that
lim q and lim q
vNvN·
and v2
....
0 as
v .... O. Thus
Note that, in evaluating
lim q,
_N
we only let v tends N but, as a rule, do not let v = N.
Indeed, sometimes N even is not in the domain of the function q = g(v). Example: q = (lv2)/(Iv). For this function, N= 1 is not in the domain of the function, and we cannot set v= 1 since it would involve division by zero. Moreover, even the limitevaluation procedure of letting vI will cuase difficulty since (lv) .... 0 as v .... 1. One way out of this difficulty is to try to transform the given ratio to a form in which v will not appear in the denominator. Since
ql ev
Iv2
(lv)(l+v)
(Iv)
l+v
(v¢ 1).
and v .... 1 implies v;r.l and (1 +v) .... 2 as v .... 1, we have
lim q 2.
_I
Example:
Find
lim q with q _ 2v +5 v_... v«1 2(v+2) +1 _ 2 + _1_ vsI vs I
since
(2v +5) v+l
and lim _1_  0, so lim q  2.
_ .. v+ 1
65
Formal View of the Limit Concept Definition of the Limit: The number L is said to be the limit of q = g(v) as v approaches N if, for every neighborhood of L, there can be found a corresponding neighborhood of N (excluding the point v=N) in the domain of the function such that, for every value of v in that Nneighborhood, its image lies in the chosen Lneighborhood. L is an open interval defined by (Lat, L+aJ Here a neighborhood of a point
=
q
{q ILaJ < q < L+~} for at > ~ > 0
q=g(v)
~~L a +
~,. '
2 ....... ,
*i i·0·~
1l..
¥"~
l' L..,. _.0_0_o.=_'"
I
~ L;_ a 1......
~
<
~'
~.
j.·i.
',
'l'
tt ~
il,;''''
6.S
Inequalities and Absolute Values
Rules of Inequalities Transitivity: a > b and b > c implies a > c a ~ b and b ~ c implies a ~ c
66
Addition and Subtraction: a>b~a±k>b±k a ze b a±k~b±k Multiplication and Division:
a>
b~
lea> kb { ka < kb
(k (k
> 0) < 0)
Squaring: a > b with b ;;::0 ~ a2 > b2 Absolute Values and Inequalities For any real number n, the absolute value of n is defined and denoted by
if n if
>0 n< 0
if nO
Thus we can write
Ix I
< n as an equivalent way n < x < n (n > 0). Also I x I
~ n if
and only if n ~ x ~ n (n ;;:: 0). The following properties characterize absolute values:
i)
ii)
iii)
[m ] + Inl ;;:: [rn + n ] [rn ] . [n ] = [m n ]
~Inl Iml n
67
Solution of an Inequality Example: Find the solution of the inequality 3x  3 > x + 1. By adding (3x) to both sides, we have 3x  3 + 3  x > x + 1 + 3  x. Thus, 2x > 4 so x > 2. Example: Solve the inequality !lx! S; 3. From !lx! 4 ~ x ~ 2, i.e. 2 S; x s 4. 4 S; x s 2. Thus 6.6 Limit Theorems
S;
3, we have 3 S lx s 3, or 
Theorems Involving a Single Equation Theorem I: If q Theorem II: If q Theorem III:
=
av + b, then lim q  aN + b
_N
=
g(v)
=
Nk
b, then
lim
_N
qb
lim v k
.... N
_
Example:
Given q = 5v + 7, then
lim q  5·2
_2
+
7  17
Example:
q
=
y3
Find
lim q
.... 2
By Theorem III, we have
lim
_2
q
23

8.
Theorems Involving Two Functions For two functions ql
=
g(v) and ~
=
h(v), if
lim q(  L( and lim q2 ~,
~N ~N
then we have
68
the following theorems: Theorem IV: Theorem V: lim (ql + q:J  LI + L2
_N
lim (qlq:J  LIL2
_N
Theorem VI:
Example:
Find
· l+v 1Im_. y..o() 2+v
Since
lim (1 +v)
y..o()

1 and
lim (2 +v)
y..o()

2,
so
1. 1v Im_
y..o()
2+v

1
2
Remark. Note that LI and L, represent finite numbers; otherwise theorems do not apply. Limit of a Polynomial Function
lim (ao +a1v
_N
+a2v2+
... + aDv D)
6.7
Continuity and Differentiability of a Function
Continuity of a Function A function q = g(v) is said to be continuous at N if lim g(v)  g(N).
_N
lim q _N
exists and
69
Thus the term continuous involves no less than three requirements: (1) the point N must be in the domain of the function; (2) lim g(v) exists; and (3)
vN
limg(v)  g(N).
vN
Remark: It is important to note that whilein discussing the limit of a functionthe point (N,L) was excluded from consideration, we are no longer excluding it in the present context. Rather, as the third requirement specifically states, the point (N,L) must be on the graph of the function before the function can be considered as continuous at point N. Polynomial and Rational Functions From the discussion of the limit of polynomial function, we know that the limit exists and equals the value of the function at N. Since N is a point in the domain of the function, we can conclude that any polynomial function is continuous in its domain. By those theorems involving two functions, we also know any rational function is continuous in its domain.
Example:
Then
lim
vN
v+ 1
2
4v
2
lim 4v2
_N
lim (v2 + 1)  N2 + 1
_N
Example: The rational function
q
is not defined at v=2 and v=2. Since v=2,2 are not in the domain, the function is discontinuous at v=2 and v=2, despite the fact that its limit exists as v + 2 or 2. Differentiability of a Function By the definition of the derivative of a function y = f(x), at Xo we know that f' ("0) exists if and only if the lim of fly/flx exists at x=x, as flx + 0, i.e., 610
f'(Xo)  lim Ay Ax
Axo()
f(Xo ....,..== lim __ + Ax) __ f(Xo)
Axo()
Ax
[differentiability
condition]
On the other hand, the function y
=
f(x) is continuous at Xo if and only if
lim f(x)  f(x';> [continuity condition].
,....,..
We want to know what is the relationship between the continuity and differentiability of a function. Now we show that the continuity of f is a necessary condition for its differentiability. But this is not sufficient. Since the notation x  Xo implies x ~ "0, so x  Xo is a nonzero number, it is permissible to write the following identity:
f(x)  f(Xo) 
f(x) f(Xo) x xo
(x  Xo)
Taking the limit of each side of the above equation as x  "0 yields the following results: Left side
=
lim (f(x)  f(Xo)  lim f(x)  f(x';>
Right side
_ lim f(x)  f(Xo) lim (x  Xo) X xo
x.
x.
 f'(Xo)[lim x  xo]
xx,
 f/(Xo)(XO x.;>  0
611
Thus
lim f(x)  f(xJ

o. So lim f(x)  f(xO> which means f(x) is continuous at x = "0.
Although differentiability implies continuity, the converse is not true. That is, continuity is a necessary, but not sufficient, condition for differentiability. The following example shows this fact. Example: f(x) =
I x I.
This function is clearly continuous at x=O. Now we show that it is not differentiable at x=O. This involves the demonstration of a disparity between the leftside and the rightside limits. Since, in considering the rightside limit X> 0, thus
lim f(x)  f(O) _ lim ~  lim 1  1
x...o·
xO
x...o·
x
x...o·
On the other hand, in considering the leftside limit, x < 0; thus
f(x)  f(O)  1m I x 1m_ 11·m_;_;:........,~ I· _ x I  I·
x...o·
xo
x...o
x
x...o·
x
 lim 1  1
x...o·
Thus,
x...o·
lim Ay
Ax
does not exist, this means the derivative of y =
y
Ix i does
not exist at x=O.
y=I%21+1
I~t___+_+1 0 1 2 3
%
4
612
VII. RULES OF DIFFERENTATION
AND THEIR USE IN COMPARATIVE STATICS
The central problem of comparativestatic analysis, that of finding a rate of change, can be identified with the problem of finding the derivative of some function y=f(x), provided only a small change in x is being considered. Before going into comparativestatic models, we begin some rules of differentiation. 7.1 Rules of Differentiation for a Function of One Variable
ConstantFunction Rule If y=f(x)=c, where c is a constant, then
dy dx
== yl == fl _ O.
Proof.
dy _ lim f(x,) f(x) _ lim c  c _ lim _ O. dx x'_ Xl x x'x X/_X x'_
We can also write
dy dx
df dx
as
d _y dx
d  _f. dx
So we may consider d/dx as an operator symbol. PowerFunction Rule If y = f(x) = xa where a is any real number  00 < a <
00,
_!f(x)  ax" '.
dx
71
Remark.
(i) If a = 0, then
d _xo dx
d  _1  0 dx
(ii) If a = 1, then y = x. Thus
::  1.
For simplicity, we prove this rule only for the case where a=n, where n is any positive integer. since
then
xl1Xci"
xxo
 xl11 + XoXI12 + ~XI13 + ... + x:;I
Thus,
f/{Xo)
_ lim f{x)  f{Xo) _ lim x 11_ Xci"
.
x......
xXo
.
xXo
+ ... + x:;I]
 lim [xl11+ XoXI12 + ~XI13
 Xo +Xo
111 111
+Xo
111
+ ... +Xo
111
 I1Xo
111
Example:
Suppose y=f'(xj=x".
Then y' = 3x4.
Example: Suppose
y  fi(x) v'x. x
Then Y I  dx x 'h  Ih x Ih . d
In particular, we can know that
£1(2) _Ih _1
{i
_.fi.
4
72
PowerFunction Rule Generalized
dy df __acx. dx dx
If the function is given by y = ex", then
a_I
Example: Suppose y=2x.
Then
Example: Suppose y=4X3. Then
dy  4'3x31 dx

12x2
Example: Suppose y = 3x2. Then
dy _ 6x 21 dx
_
6x 3.
7.2 Rules of Differentiation Involving Two or More Functions of the Same Variable Let f(x) and g(x) be two differentiable functions. We have the following rules: SumDifference Rule
d _ [f(x) dx
± g(x)]   d f(x)!  d g(x)  f/(X) ± g'(x)
dx dx
This rule can easily be extended to more functions
d [DL (x) ] dx
iI

D D L d fi(x)  L dx
iI iI
f;(x).
Example:
~[ax2+bx+c] dx
 Zax+ b
Example: Suppose a shortrun totalcost function is given by c = Q3  4Q2 + lOQ + 75. Then the marginalcost function is the limit of the quotient t:.CIt:.Q, or the derivative of the C function:
73
de _ 3Q2 _ 4Q dQ
+
10.
In general, if a primitive function y=f(x) represents a total function, then the derivative function dy/dx is its marginal function. Since the derivative of a function is the slope of its curve, the marginal function should show the slope of the curve of the total function at each point x. Sometimes, we say a function is smooth if its derivative is continuous. Product Rule
~ [f(x)g(x)]  f(x) ~ g(x) + g(x) ~ f(x) dx dx dx  f(x)g'(x) + g(x)f'(x)
Proof.
~ [f(xJg(xJ] dx

 lim f(x)g(x)  f(xJg(xo) xx, xXo xxo xXo
+
1
. Im f(x)g(x)  f(x)g(xo) + f(x)g(xJ  f(xJg(xo) _
_ lim f(x)[g(x)  g(xJ] +g(xJ[f(x)  f(xJ]
xx,
· ti() g(x)  g(xJ 1 im x xx, xxo
1· (' [f(x)  f(xJ] img x()l xx, xXo
 f(xJg'(xJ
+
g(xJf'(xJ
Since this is true for any x = Xo, this proves the rule. Example: g' (x) = 6x. Suppose y = (2x Hence,
+
3)(3x2).
Let f(x) = 2x
+
3 and g(x) = 3x2• Then f'(x) = 2
~[(2x+3)(3x~] dx  18x2 + 18x
 (2x+3)6x+3x2·2
 12x2 + 18x + 6x2
74
As an extension of the rule to the case of three functions, we have
~ [f(x)g(x)h(x)]  f'(x)g(x)h(x)
dx
+
f(x)g'(x)h(x)
+ f(x)g(x)h'(x).
Finding MarginalRevenue Function from AverageRevenue Function Suppose the averagerevenue (AR) function is specified by AR = 15  Q. Then the totalrevenue (TR) function is TR
Ei
AR . Q = 15Q _ Q2.
Thus, the marginalrevenue (MR) function is given by
MR •
~
dQ
TR  15  2Q.
In general, if AR
=
f(Q), then
TR == AR . Q = Qf(Q). Thus
MR
!!IE
~
dQ
TR  f(Q)
+ Qf'(Q).
From this, we can tell relationship between MR and AR. Since MR  AR
=
[f(Q) + Qf'(Q)]  f(Q)
=
Qf'(Q),
they will always differ the amount of Qf'(Q). Also, since
75
AR
&
Q  Q  P,
TR
PQ
we can view AR as the inverse demand function for the product of the firm. If the market is perfectly competitive, i.e., the firm takes the price as given, then P=f(Q)=constant. Hence f' (Q) = O. Thus MR  AR = 0 or MR = AR. Under imperfect competition, on the other hand, the AR curve is normally downwardsloping, so that f'(Q) < O. Thus MR < AR. Quotient Rule
~ [ f(x) ] dx g(x)
_ f'(x)g(x)  f(x)g'(x) g2(X)
We will come back to prove this rule after learning the chain rule. Example:
(i)
~ [2X3] dx x+l
_ 2(x+l)(2x3)(I) (x+l? 5 (x + 1)2
5(x2+1)5x(2x) (x2+1)2
...) (111  d dx
[ax2+bJ

ex

.....:=..:.....,,'
2ax(ex) c(ax2

(ex)2 b)
(ax +b)e
2
ax2 ~

b
(CX)2
Relationship Between MarginalCost and AverageCost Functions As an economic application of the quotient rule, let us consider the rate of change of average cost when output varies. Given a total cost function C = CCQ),the average cost (AC) function and the marginalcost (MC) function are given by 76
AC
Iii
C(Q)
~
(Q> 0)
and MC
=
C'(Q). AC:
The rate of change of AC with respect to Q can be found by differentiating
From this it follows that, for Q
> O. >
~ AC  0 iff MC(Q)  AC(Q). dQ < <
>
100_._
90'"'T" 8070
MC
= 3Q2
24Q
+ 60
~ 5040302010
e 8
60
~
I
I 
o
t 
I
~t.!4~j~f19f110
Q
7.3 Rules of Differentiation Involving Functions of Different Variables
Now we consider cases where there are two or more differentiable which has a distinct independent variable. functions, each of
77
Chain Rule
If we have a function z=f(y), where y is in turn a function of another variable x, say,
y=g(x), then the derivative of z with respect to x is given by
dz dz dy _  _ . _  f' (y)g '(x). dx dy dx
[Chain Rule]
The chain rule appeals easily to intuition. Given a ill{, there must result a corresponding Ll.y via the function y=g(x), but this Ll.ywill in turn being about a Ll.z via the function z=f(y). Proof. Note that
dz _ lim az _ lim az . dx AxoQ ax AxoQ ay ax
ez.

Since
ill{ ......0
implies Ll.y...... we have 0,
dz dx
dz • dy _ f'(y)g'(x). dy dx
Q.E.D.
In view of the function y= g(x), we can express the function z=f(y) as z=f(g(x», where the contiguous appearance of the two function symbols f and g indicates that this is a compose function (function of a function). So sometimes, the chain rule is also called the compositefunction rule. As an application of this rule, we use it to prove the quotient rule. Let Z=y·l and y=g(x).
dz dx
Then z = g.l(X) = l/g(x) and
g(x) g2(X)
dz dy 1 _ . _   _g(x) dy dx y2
Thus,
78
~ [ f(x) ] dx g(x)
_ ~ [f(x) . gl(X)] dx
_ f'(X)gl(X) + f(x) ~ [gl(X)] dx _ f'(X)gl(X) + f(x) f'(x)g(x)  f(x)g'(x) g2(X) [_ g(x) ) g2(x)
Q.E.D.
Example:
If z = 3y2 and y
=
2x + 5, then
dz _ dz dx dy
dy _ 6y(2) _ 12y _ 12(2x + 5) dz
Example:
If z
=
y  3 and y
=
y
=
x', then
dz _ dz dy _ 1 . 3x2 _ 3x2 dx dy dz
Example:
z
=
(x2
+
3x  2)17. Let z
=
y17 and y
=
x2
+
3x  2.
Then
dz _ dz dy _ 17y16 dx dy dx  17(x2+3x2)(2x+3).
•
(2x+3)
Example:
Suppose TR
=
f(Q), where output Q is a function of labor input L, or Q is
=
g(L).
Then, by the chain rule, the marginal product of labor (MPJ
MP
L
_ dR _ dR dL dQ
dQ _ f'(Q) '(L) dL g
 MR . MPL
Thus the result shown above constitutes the mathematical economics that
statement of the wellknown result in
79
InverseFunction Rule
If the function y =f'(x) represents a onetoone mapping, i.e., if the function is such that
a different value of x will always yield a different value of y, then function f will have an inverse function x = f'(y). Here, the symbol f' is a function symbol which signifies a function related to the function f; it does not mean the reciprocal of the function f(x). When x and y refer specifically to numbers, the property of onetoone mapping is seen to be unique to the class of function known as monotonic function. A function f is said to be monotonically increasing (decreasing) if XI > X2 implies f(x,) > f(xz) [f(x,) < f(xz)]. In either of these cases, an inverse function f' exists. A practical way of ascertaining the monotonicity of a given function y=f(x) is to check whether the f'(x) always adheres to the same algebraic sign for all values. Geometrically, this means that its slope is either always upward or always downward. Example: Suppose y = 5x + 25. Since y' = 5 for all x, the function is monotonic and thus the inverse function exists. In fact, it is given by x = 1/5 Y  5. Generally speaking, if an inverse function exists, the original and the inverse functions must be both monotonic. Moreover, if f' is the inverse function of f, then f must be the inverse function of fl. For inverse functions, the rule of differentiation is
Proof.
dx  1· _ 1m .1X  1° _ 1  _~_ 1 1m dy >Y. ~ >Y. ~ lim_.1y .1~ X_'" .1x

1
yl
Q.E.D.
Example:
Suppose y =
X
S
+
x. Since y' = 5x4
+
1, so
dx dy
1 5x4+ 1
710
7.4 Partial Differentiation So far, we have considered only the derivative of functions of a single independent variable. In comparativestatic analysis, however, we are likely to encounter the situation in which several parameters appear in a model, so that the equilibrium value of each endogeneous variable may be a function of more than one parameter. Because of this, we now consider the derivative of a function of more than one variable. Partial Derivatives Consider a function
where the variables xj(i= 1,2, ... ,n) are all independent of one another, so that each can vary by itself without affecting the others. If the variable X; changes Llxj while the other variables remain fixed, there will be a corresponding change in y, namely, fly. The difference quotient in this case can be expressed as
!ly f(xl' X2' ..., Xi)' x; +!lxp Xi+l' ..., xD) !lx; !lx;

f(xl' ..., xD)
If we take the limit of fly/Llxj, that limit will constitute a derivative.
We call it the partial
derivative of y with respect to x., The process of taking partial derivatives is called partial differentiation. Denote the partial derivative of y with respect to X;by
oy
ax;'
i.e.
Also we can use fj to denote 8y/8X;. If the function happens to be written in terms of unsubscripted variables, such as y=f(u,v,w), derivatives. one also uses, t:" fv, fw to denote the partial
711
Techniques of Partial Differentiation Partial differentiation differs from the previously discussed differentiation primarily in that we must hold the other independent variables constant while allowing one variable to vary.
ay
aXI
=
ay
a~

2  6Xl + ~ aXI 
af
af a~
2  Xl + 8X2
Example: For y = f(u,v) = (u+4)(3u+2v),
we have
~~
!!!
fu  (3u+2v) +(u+4)'3
 6u+2v+12  2(3u+v +6)
av •
ay
fv  2(u+4)
When u=2 and v= 1, then fu(2,1) = 2x13 = 26 and fv(2,1) = 2x6 = 12. Example: Given y = (3u  2v)/(u2 + 3v),
ay au
ay av

3(u2+ 3v)  (3u 2v)(2u) (u2+ 3V)2 2(u2 + 3v)  (3u  2v)'3 (u2+3v)2
3u2 + 4uv + 9v (u2+3v)2
u(2u+9) (u2+ 3V)2
712
7.5 Applications to ComparativeStatic
Analysis we can at last tackle value of an
Equipped with the knowledge of the various rules of differentiation, the problem posed in comparativestatic parameters. Market Model For the onecommodity market model: (a,b (c,d
analysis: namely, how the equilibrium
endogeneous variable will change when there is a change in any of the exogeneous variables or
QcI = a  bp
Q.
= c + dp
> 0)
>
0),
the equilibrium price and quantity are given by

a+c p_ h+d
Q

.;:;:a
adbe
These solutions will be referred to as being in the reduced form: the two endogeneous variables have been reduced to explicit expressions of the four independent parameters, a, b, c, and d.
To find how an infinitesmal change in one of the parameters will affect the value of P or Q, one has only to find out its partial derivatives. If the sign of a partial derivative can be determined, constitutes ascertained, we will know the direction in which P will move when a parameter a qualitative conclusion. If the magnitude conclusion. of the partial say, it will constitute a quantitative changes; this can be a clear The latter derivative
To avoid misunderstanding,
distinction should be made between the two derivatives, the supply function. The derivative
oQloa and oQioa.
derivative is a concept appropriate to the demand function taken alone, and without regard to
oQloa, on the other hand, to the equilibrium quantity which
takes into account interaction of demand and supply together. To emphasize this distinction, we refer to the partial derivatives of P and Q with respect to the parameters as comparativestatic derivatives.
713
For instance, for P, we have
oP
ob   (b+d)l
(a+c)
oP
od
(a+c) (b+ dj?
Thus
oP _ oP oa oc


>
0 and OP _ oP

ob
od
< o.
Q
(Increase in a)
I !
(Increase
in
b)
a
D
0
pp.
Q
:
P
0
P'
(b)
P
(a)
(Increase m e)
Q
I
(Increase
in
d)/ S'
/
D
/
D
P p'
(e)
P
p'
P
(d)
714
NationalIncome y T
Model (equilibrium condition) (a (y
= =
C + 10 + Go
'Y
C=a
+ (3(YT) + oY
> 0; 0 < > 0; 0 <
(3 0
< 1)
<
1) C, and taxes T.
where three endogeneous
variables are the national income Y, consumption
The equilibrium income (in reduced form) is
y_
Thus,
a{ji'+lo
+Go
l{3+{jo
1 >0 1 {j + {jo
8Y 8i'
[the governmentexpenditure multiplier]

aF
8Y

{j 1 {j + {jo
<0
{j(a  (ji' +10 +G) (1{j+{j0)2
<0
7.6 Note on Jacobian Determinants
Partial derivatives can also provide a means of testing whether there exists functional This is related to
(linear or nonlinear) dependence among a set of n functions in n variables. the notion of Jacobian determinants. Consider n differentiable functions in n variables not necessary linear
YI
Y2
=
=
f(x"x2,···,xj f(x,,~, ...,xj
where the symbol
r denotes
the ith function, we can derive a total of n2 partial derivatives.
715
(il,2, ...n; j 1,2, ...n). , ,
We can arrange them into a square matrix, called a Jacobian matrix and denoted by J, and then take its determinant, the result will be what is known as a Jacobian determinant (or a Jacobian, for short), denoted by
IJ I:
c3y1
axl
c3y1 ~ c3y2 ~ c3y"
c3yl
ax"
c3y2
IJI •
c3(y1'Y2""'Y,)
c3(xI~"",x,)
• axl
c3y2
ox"
c3y"
c3y"
axl ax2
ax"
Example:
Consider two functions:
YI
Y2
= =
2XI + 3x2 4xi + 12xIX2 + 9x~
Then the Jacobian is
c3yl
IJI •
axl ax2
c3y2
c3yl 2
(8xI +1~)
3 (12x1 + 18x2)
ax
c3y2 ~
1
A Jacobian test for the existence of functional dependence among a set of n functions is provided by the following theorem:
Theorem:
The Jacobian I J I defined above will be identically zero for all values of x., X2,··· .x, if and only if the n functions fl,f, ... ,t" are functionally (linear or nonlinear) dependent.
For the above example, since
716
IJI 
for all XI and X2, y1 and y2 are functionally dependent. Example: Consider the linearequation system: Ax
In fact, Y2is simply YI squared.
=
d, i.e.,
aUXl ~IXl
+ +
a12x2 + ~X2
+
+ +
a1nx" = d, ~x"
=
d2
We know that the rows of the coefficient matrix A are linearly dependent if and only if
=
IA I
O. This result can now be interpreted as a special application of the Jacobian criterion of
functional dependence. Take the left side of each equation Ax this, the elements of
=
d as a separate function of the n variables
Xt>X2, ,Xn, and denote these functions by Yt>Y2, ,Yn. Then we have oY/OXj ... ...
IJ I will
be precisely the elements of A, l.e.,
= 3;,;. In view of 11 I = I A I and thus the
Jacobian criterion of functional dependence among Yt>Y2, ,Ynis equivalent to the criterion ... = 0 in the present linear case.
IA I
717
VIII.
COMPARATIVESTATIC ANALYSIS OF GENERALFUNCTIONS
The study of partial derivatives has enabled us, in the preceding chapter, to handle the simple type of comparativestatic problems, in which the equilibrium solution of the model can be explicitly stated in the reduced form. We note that the definition of the partial derivative requires the absence of any functional relationship among the independent variables. As applied to comparativestatic analysis, this means that parameters and/or exogeneous variables which appear in the reducedform solution must be mutually independent. However, no such expediency should be expected when, owing to the inclusion of general functions in a model, no explicit reducedform solution can be obtained. In such a case, we will have to find the comparativestatic derivatives directly from the originally given equations in the model. Take, for instance, a simple nationalincome model with two endogeneous variables Y and C:
Y
C
= =
C + 10 + Go
C(y,To)
[equilibrium condition] [To: exogeneous taxes]
which reduces to a single equation
to be solved for Y. We must, therefore, find the comparativestatic derivatives directly from this equation. How might we approach the problem? Let us suppose that an equilibrium solution Y does exist. We may write the equation


even though we are unable to determine explicitly the form which this function takes. Furthermore, in some neighborhood of Y, the following identical equality will hold: Y
EI
C(y,To) + 10 + Go.

Since Y is a function of To, the two arguments of the C function are not independent. To can in this case affect C not only directly, but also indirectly via Y. Consequently, partial differentiation is no longer appropriate for our purposes. In this case, we must resort to total differentiation (as against partial differentiation). The process of total differentiation can lead 81
us to the related concept of total derivative. Once we become familiar with these concepts, we shall be able to deal with functions whose arguments are not all independent so that we can study the comparativestatics of a generalfunction model. 8.1 Differentials
The symbol dy/dx has been regarded as a single entity. We shall now reinterpret as a ratio of two quantities, dy and dx. Differentials and Derivatives Given a function y=f(x), we can use the difference quotient t>.y/Axto represent the ratio of change of y with respect to x. Since (8.1)
Ay
== [:~]
Ax
the magnitude of t>.ycan be found, once the t>.y/Axand the variation Ax are known. If we denote the infinitesimal changes in x and y, respectively, by dx and dy, the identity (8.1) becomes (8.2)
dy
liE [:~]
dx
or
dy  f/(X)dx
The symbols dy and dx are called the differentials of y and x, respectively. Dividing the two identities in (8.2) throughout by dx, we have
(dy)
(dx)
=
[dY] dx
or
(dy) _ f'(x) (dx)
This result shows that the derivative dy/dx == f'(x) may be interpreted as the quotient of two separate differentials dy and dx. On the basis of (8.2), once we are given f'(x), dy can immediately be written as f'(x)dx. The derivative f' (x) may thus be viewed as a "converter" that serves to convert an infinitesimal change dx into a corresponding change dy.
82
Example: is
Given y = 3x2 + 7x  5, find dy. Since f' (x) = 6x + 7, the desired differential
dy = (6x + 7)dx Remark. The process of finding the differential dy is called differentiation. Recall that we have used this term as a synonym for derivation. To avoid confusion, the word "differentiation" with the phrase "with respect to x" when we take the derivative dy/dx. The following diagram shows the relationship between "Ay" and "dy"
y
I
.Y=/(%)
o ~._
%
tJ..y _ [tJ..y] tJ..x
dy _ [dY] dx
~
~
_ CB AC _ CB
AC _ CD AC _ CD AC
which differs from Ay by an error of DB. Differentials and Point Elasticity As an illustration of the application of differentials in economics, let us consider the elasticity of a function. For a demand function Q=f(p), for instance, the elasticity is defined as (AQ/Q)/(AP/P). Now if AP  0, the AP and AQ will reduce to the differential dP and dQ, and the elasticity becomes
83
Ed
E
dQ/Q dQ/dP marginal demand function dPIQ Q/P  ave=ra:""ge:'deman:d"""fun""ct""ion
In general, for a given function y=f(x),
E
)'X
the point elasticity of y with respect to x as


dy/dx y/x
Ed

""'=;":::.average function
Since dQ/dP = 2 and Q/P =
marginal
function
Example: Find
(1002P)/2, unit elastic so
Ed
if the demand function is Q = 100  2P. and elastic for 25
(I Ed I
= (P)/(50P). = 1) for P=25,
Thus the demand is inelastic (
I Ed I <
1) for 0
<
P
<
25,
< P < 50.
8.2
Total Differentials
The concept of differentials can easily be extended to a function of two or more
independent variables. S = S(y,i)
Consider a saving function
where S is savings, Y is national income, and i is interest rate.. If the function is continuous and possesses continuous partial derivatives, dS  ~
8Y
the total differential
is defined by
dY
+
as di
ai
That is, the infinitesimal infinitesimal change in i. Remark.
change in S is the sum of the infinitesimal
change in Y and the
If i remains constant, the total differential will reduce to a partial differential:
ay Furthermore,
as
[
dS ] dY
i cxmIWII
general case of a function of n independent variables y=f(xl>x1, ... ,xJ. the total
differential of this function is given by
in which each term on the right side indicates the amount of change in y resulting from an infinitesimal change in one of the independent variables.
84
Similar to the case of one variable, the n partial elasticities can be written as
Ea,  
of
OXi f

Xi
(i 1,2,...n). ,
8.3
Rules of Differentials
Let c be constant and u and v be two functions of the variables Xl,X2""'x". Then the following rules are valid: Rule I. de = 0 Rule II. dfcu") = caua1du Rule III. d(u ± v) = du ± dv Rule IV. d(uv) = vdu + udv Rule V. d(u/v) = lIv2 (vdu  udv) Example: Find dy of the function y = 5x~ + 3x2. There are two ways to find dy. One is the straightforward method by finding aflax1 and aflax2: af/ax1 = IOxl and aflax2 = 3, which will then enable us to write
The other way is to use the rules given above by letting u=Sx] and V=3X2: dy = d(5xD + d(3x~ = IOx1dx1+ 3dx2 [by Rule III] [by Rule II].
Example: Find dy of the function y = 3x~ + X1X~.Since fl = 6x1 + x~ and f2 = 2X1X2,the desired differential is
By applying the given rules, the same result can be arrived at dy = d(3xD + d(xlx~ = 6x1dx1 + x~dxl + 2x1x2dx2 = (6x1 + x~dxl + 2x1x2dx2
85
Example: For the function
(XI +2x2) fl  """32xI
then
y   (XI +2xJ dXI + _!_2 dx, 2x 2XI
3 1
The same result can also be obtained by applying the given rules:
dy 4
1
4xI
[2xld(xl +X~ (XI +xJd(2xl)]
2
2
[by Rule V]
~ [2x~(dxl +dx~ (XI +x2)4xldxl] 4xI
4
1
[2XI(XI+2xJdxl+2xldx2]
2
4xI (XI +2x2) 23 dXI XI
+
1
2
2xI
dx,
For the case of more than two functions, we have Rule VI. d(u ± v ± w) = du ± dv ± dw Rule VII. d(uvw) = vwdu + uwdv + uvdw. 8.4 Total Derivatives Consider any function y = f(x,w) where x=g(w). Unlike a partial derivative, a total derivative does not require the argument x to main constant as w varies, and can thus allow for the postulated relationship between the two variables. The variable w can affect y through two channels: (1) indirectly, via the function g and then f, and (2) directly, via the function. Whereas the partial derivative fw is adequate for expressing the direct effect alone, a total derivative is needed to express both effects jointly.
86
To get the total derivative,
we first get the total differential
dy
=
fxdx
+
fwdw.
Dividing both sides of this equation by dw, we have the total derivative
dy _ f dx dw x dw
+
f
W
dw dw
+ox dw
Example: y
oy
dx
ow
oy
Find the dy/dw, given the function
=
f(x,w)
=
3x +
w where
g(w)
=
2w2
+
w
+
+ 4.
3.
dy  f _ dx _ dw x dw
f
W
 3(4W +1)  2w  lOw
As a check, we may substitute the function g into f, to get y = 3(2w
+
w
+
4)  w2 = 5w2 Then
+
3w
+
12
which is now a function of w alone.
dy _ lOw + 3 dw
the identical answer.
Example:
Find du/dc, given u = u(c,s) with s=g(c).
Thend
dUe
ou oc
+
og(c)
ou
g'(e)
.
A Variation on the Theme For the function
with
XI
= g(w) and
dy dw
X2
= hew), the total derivative of y is given by
+
of
oX2

dx, dw
+.
ow
of
87
Example:
Q
Let the production function be
=
Q(K,L,t)
where K is the capital input, L is the labor input, and t is the time which indicates that the production function can shift over time in reflection of technological change. Since capital and labor can also change over time, we may write K
=
K(t) and L
=
L(t).
Thus the rate of output with respect to time can be denoted as dQ _dt
__
oQ dK oK dt
+_+_dL
oQ
oQ
ot
oL dt
Another Variation on the Theme Now if a function is given,
with Xl = g(u,v) and ~ = h(u,v), we can find the total derivative of y with respect to u (while v is held constant). Since dy  _of dx, ox,
+
oX2
of dX + of du 2
OU
+_
OV
of dv.
dividing both sides of the above equation by du, we have dy du
oy dx,  __ ox, du oy dx,  __ ox, du
oy du oy dv au du + du ov oy dx, oy [dV G si + __ . +du  smce v IS constant] o~ du ou
+ __
oy
dx,
oX2 du
+ 
Since v is held constant, the above is the partial total derivative, we redenote the above equation by the following notation:
+ __
oy oX2 +oy o~ ou ou
Remark. In the cases we have discussed, the total derivative formulas can be regarded as
88
expressions of the chain rule, or the compositefunction rule. Also the chain of derivatives does not have to be limited to only two "links"; the concept of the total derivative should be extendible to cases where there are three or more links in the composite function. 8.S Derivatives of Implicit Functions The concept of total differentials can also enable us to find the derivatives of the socalled "implicit functions." Implicit Functions A function given in the form of y = f(xhx2,... ,x,.) is called an explicit function, because the variable y is explicitly expressed as a function of x. But in many cases y is not an explicit function of X1>X..., .x.; instead, the relationship between y and x.,... .x, is given with the form 2 of
Such an equation may also be defined as implicit function y = f(xhx2' ... .x.), Note that an explicit function y = f(xh ... .x.) can always be transformed into an equation F(y,xh ... ,xj yf(xhx2' ... 'x,.) = O. The reverse transformation is not always possible. In view of this uncertainty, we have to impose a certain condition under which we can be sure that a given equation F(y,xh ... ,xj = 0 does indeed define an implicit function y=f(x1,x2, ... ,x,.). Such a result is given by the socalled "implicitfunction theorem." ImplicitFunction Theorem. Given F(y,XhX2'... 'x,.) = 0, if (a) the function F has continuous partial derivatives Fy,FX"F ...,Fx.' .., and if (b) at a point (Yo,xlO, ,x"o) satisfying ... F(yo,xlO, ,x"o) = 0, Fy is nonzero, then there exists an ndimensional neighborhood of ... (XIO' 'XoO)' , in which y is an implicitly defined function of variables x., ... ,xn, in the form of ... N y=f(x., ...'x.,), and F(y,xh ... .x.) = 0 for all points in N. Moreover, the implicit function f is continuous, and has continuous partial derivatives fh ... ,fn• Derivatives of Implicit Functions Differentiating F, we have dF=O, or
89
Suppose that only y and Xl are allowed to vary. Then the above equation reduce to Fydy + ,Fjdy =
o.
Thus
dy
dxl
L
vuiabIe
coastmt
·
!1.. axl
Fy
FI
(i  1,2,...,n)
In the simple case where the given equation is F(y ,x) = 0, the rule gives
Example:
Suppose y  3x4 = O. Then

dy dx
~
F Fy
  1
12x3
 12x
3
In this particular case, we can easily solve the given equation for y, to get y=3x4 so that dy/dx
=
12x3• F(x,y) = x2 +
dy dx
Example:
f 9
2x 2y
= O. Thus,
x y
(y;eO)
Example:
For F(y ,x, w) =
tx2
+ w3 + yxw  3 = 0, we have
ay
ax
In particular, at point (1,1,1), oy/ox = 3/4. Example: Q=f(K,L). Assume that the equation F(Q,K,L) = 0 implicitly defines a production function Then we can find MPK and MPL as follows:
and
In particular, we can also find the MRTSLK which is given by
810
Extension to the SimultaneousEquation Case Consider a set of simultaneous equations. F1(Yt> F2(Yl' ,YD; x., 'YD; X1,
,x",) ,x",)
=
=
0 0
Suppose Fl,F2, ... ,P are differentiable. Then we have
+ ... +
aFt _dx ]
aXm m
+ ... +
aF _dx
2
ax
] m
m
.................................................................................
Or in matrix form,
aFt aFt oFi ... aYt OJ2 OJ" aFl aFl ... aFl aYt OJ2 OJ" of'' of'' ... of" aYt aY2 OJ"
dYt dY2
dy"
aFt aFt ... aXt aX2 aFl aFl ... aXt oX2 of" of" OXt aX2
...
oFt aXm aFl aXm aF" aXm
dxt dx2 dxm
If we want to obtain partial derivatives with respect to
Xj,
we can let dx.s=O for t¢ i. Thus, we
have the following equation:
811
aFI aYI ap ay. aFn ay.
aFI ... aFI ityn ity2 ap ity2
aYI ax, aY2 ax, aYn ax,
aFI ax, ap ax, aFn ax,
...
aP ityn
aFn aFn ... ayz ityn
Now suppose the following Jacobian determinant aFI aFI oF l
is nonzero: of·
Oy. Oy.
Oyl Oyl
Oy"
oFl
111
.1 O(FI,Fl, ...
,F,,)
1
OF2 
iJ(y.'Yl'···'Y")
Oy"
'" 0

of''
Oy.

of''
Oyl

of''
Oy"
Then, by Cramer's
rule, we have
(j 1,2, ... n, i 1,2, ... , ,m).
where
IJ I is obtained
j i
by replacing the jth column of
IJ I with
aFl, [ aXj
aF2, ...,
8xj
8Fn]'.
8xj
Of course, we can find these derivatives by inversing the Jacobian matrix J:
812
c1y1
ox,
c1y2
ax, oY" ax,
oFI oFI ... c1y1 °Y2 oFl oFl ... aY2 c1y1
oFI OY" oFl oY"
I
oFI ox, oFl ax, of" ax,
of" of" ... of" aYI c1y2 ay"
In the compact notation,
Example:
Let the nationalincome
model be rewritten in the form:
y  c  10 T  'Y  oY Since
Go = 0 C  ex  (J(Y  T) = 0
=
0
of' of' of' or ec oT oF'l oF'l oF2 iJi  or ec oT oF oF oF or sc oT
3 3 3

1
I 0
p 1 P  IIhp~. ~ 0 1
Suppose all exogeneous variables and parameters
are fixed except Go, then we have
[
{3 0
1 1 0]
1 {J 0 1
[ay/aGo
a =/0 Go
enso o
1 [ 1]
0 0
so that we can solve the above equation for, say, ay laGo which comes out to be

813
1 1 0
o o
1 0
P
1
1
IJI
8.6 Comparative Statics of GeneralFunction Models Consider a singlecommodity market model:
Qd Qd Q.
= Q. = D(p,Y = S(P)
o)
[equilibrium condition] [8D18P < 0; 8DI8Yo > 0]
[dS/dP
> 0]
where Yo is an exogeneously determined income. From this model, we can obtain a single equation: D(p,Yo)

S(P)
=
O.
Even though this equation cannot be solved explicitly for the equilibrium price P, by the implicitfunction theorem, we know that there exists the equilibrium price P which is the function of Yo:

such that D(p,Yo)


S(P)

E
O.
It then requires only a straight application of the implicitfunction rule to produce the comparativestatic derivative, dP/dYo:
dP dYo

(JDI(JP dS/dP
>0
Since Q = S(P), thus we have
dQ dYo



dS dP >0 dP dY o

814
IX. OPI'IMIZATION:
A SPECIAL VARIETY OF EQUILIBRIUM
ANALYSIS
From now on, our attention will be turned to the study of goal equilibrium, in which the equilibrium state is defined as the optimal position for a given economic unit and in which the said economic unit will be deliberately striving for attainment of that equilibrium. Our primary focus will be on the classical techniques for locating optimal positions  those using differential calculus. 9.1 Optimal Values and Extreme Values Economics is by large a science of choice. When an economic project is to be carried out, there are normally a number of alternative ways of accomplishing it. One (or more) of these alternatives will, however, be more desirable than others from the standpoint of some criterion, and it is the essence of the optimization problem to choose. The most common criterion of choice among alternatives in economics is the goal of maximizing something or of minimizing something. Economically, we may categorize such maximization and minimization problems under general heading of optimization. From a purely mathematical point of view, the collective term for maximum and minimum is the more matteroffact designation extremum, meaning an extreme value. In formulating an optimization problem, the first order of business is to delineate an objective function in which the dependent variable represents the object of maximization or minimization and in which the set of independent variables indicates the objects whose magnitudes the economic unit in question can pick and choose. We shall therefore refer to the independent variables as choice variables. 9.2 Relative Maximum and Minimum: FirstDerivative Test Consider a generalform objective function y=f(x).
y
Three specific cases of functions
B
C
o
(0)
'_ (b)
.. _%
0
% (c)
91
Remark. The points E and F in (c) are relative (or local) extremum, in the sense that, each of these points represents an extremum in some neighborhood of the point only. We shall continue our discussion mainly with reference to the search for relative extrema. Since an absolute (or global) maximum must be either a relative maxima or one of the ends of the function. Thus, if we know all the relative maxima, it is necessary only to select the largest of these and compare it with the end points in order to determine the absolute maxi~um. Hereafter, the extreme values considered will be relative or local ones, unless indicated otherwise. FirstDeriyative Test Given a function y=f(x), the first derivative f'(x) plays a major role in our search for its extreme values. For smooth functions, relative extreme values can only occur where f'(x) = 0, which is a necessary (but not sufficient) condition for a relative extremum (either maximum or minimum).
~y
y
b
!l" tz
;.'cO:
D
A
i'
(a) (b) (c)
i
".
f
/y=f(x)
,
B
0
X2
f!!'
Xl
(a)
(b)
Firstderivative test relative extremum. If f' (Xo) = 0, then the value of the function at~, f(Xo), will be A relative maximum if f'(x) changes its sign from positive to negative from the immediate left of the point Xo to its immediate right. A relative minimum if f'(x) changes its sign from negative to positive from the immediate left of Xo to its immediate right. No extreme points if f'(x) has the same sign on some neighborhood. y = (xI)", x= 1 is not an extreme point even f'(1)=O. 92
Example;
Example; y = f(x) = x3  12x2 + 36x + 8 Since f'(x) = 3x2  24 + 36, to get the critical values, i.e., the values of x satisfying the condition f'(x) = 0, we set f'(x) = 0, and thus 3x2  24x + 36 = O. Its roots are x, = 2 and X2= 6. It is easy to verify that f' (x) > 0 for x < 2, and f' (x) < 0 for x>2. Thus x=2 is a maximum point and the corresponding maximum value of the function f(2) =40. Similarly, we can verify that x=6 is a minimum point and f(6) = 8. Example; Find the relative extremum of the averagecost function


AC = f(Q) = Q2  5Q + 8. Since f'(2.5)=0, extreme point. f'(Q) < 0 for Q < 2.5, and f'(Q) > 0 for Q > 2.5, so Q = 2.5 is an

9.3 Second and Higher Derivatives Since the first derivative f'(x) of a function y=f(x) is also a function of x, we can consider the derivative of f'(x), which is called second derivative. Similarly, we can find derivatives of even higher orders. These will enable us to develop alternative criteria for locating the relative extrema of a function. The second derivative of the function f is denoted by f"(x) or dZy/dx2. If the second derivative f"(x) exists for all x values, f(x) is said to be twice differentiable; if, in addition, f" (x) is continuous, f(x) is said to be twice continuously differentiable. The higherorder derivatives of f(x) can be similarly obtained and symbolized along the same line as the second derivative: f'" (x), f4)(X), ... , fD)(x) or
Remark. d'y/dx" can be also written as (dD/dxD)y, here the dD/dx part serves as an operator w symbol instructing us to take the nth derivative with respect to x.
D
93
Example: Then
y = f(x) = 4X4 x3 + 17x2 + 3x  1
f'(x) = 16x3  3x2 + 34x + 3 f"(x) = 48x2  6x + 34 fll'(X) = 96x  6 f<4)(X) 96 = f<Sl(x) 0 =
Example.
Find the first four derivatives of the function
y  g(x)  _ x l+x
(X;C
1)
g'(x) = (1 +X)2
g"(X)
= 2(1 +xy3 g"'(X) = 6(1 +x)4 g(4)(X)= 24(1 +x)?
Remark. A negative second derivative is consistently reflected in an.inverse Ushaped curve; a positive second derivative is reflected in an V shaped curve. 9.4 SecondDerivative Test Secondderivative test for relative extremum. If f'(Xo) = 0, then the value of the function at Xo,f(Xo)will be (a)
(b)
A relative maximum if f" (Xo) < 0 A relative minimum if fll(Xo) > O.
This test is in general more convenient to use than the firstderivative test, since it does not require us to check the derivative sign to both the left and right of x. Example: y = f(x) = 4x2  x. Since f'(x) = 8x  1 and fll(X) = 8, we know f(x) reaches its minimum at x = 118. Indeed, since the function plots as a Ifshaped curve, the relative minimum is also the absolute minimum. 94
Example; y = g(x) = Then y' = g'(x) values ~1 = 0 and ~2 = because g'(O) < 0) and
x3  3x2 + 2. = 3x2  6x and y" = 6x  6. Setting g'(x) = 0, we obtain the critical 2, which in turn yield the two stationary values g(0)=2 (a maximum g(2)=2 (a minimum because g'(2) > 0).
Remark. Note that when f'(Xo)=O, fll(Xo) < 0 (f"(Xo) > 0) is a sufficient condition for a relative maximum (minimum) but not a necessary condition. However, the condition f"(Xo) S;
o (f"(Xo)
~
0) is a necessary condition (even though not sufficient) for a relative maximum
(minimum). Condition for Profit Maximization Let R=R(Q) be the totalrevenue function and let C=C(Q) be the totalcost function, where Q is the level of output. The profit function is then given by 7r = 7r(Q) = R(Q)  C(Q). To find the profitmaximizing output level we need to find Q such that 7r'(Q) = R'(Q)  C'(Q) = 0 or R'(Q) = C'(Q), or MR(Q) = MC(Q). To be sure the firstorder condition leads to a maximum, we require







~/(Q)

 R"(Q)

 C"(Q)

<
o.
Economically, this would mean that, if the rate of change of MR is less than the rate of change of MC at Q, then that output Q will maximize profit. Example; Let R(Q) = 1200Q  2Q2 and C(Q) = Q3  61.25Q2 + 1528.5Q + 2000. Then the profit function is 7r'(Q) = _Q3 + 59.25Q2  328.5Q  2000.
95
Setting 1("1(Q) _3Q2 + 118.5Q  328.5 = 0, we have Ql = 3 and Q2 = 36.5. Since 1(""(3) = = 18 + 118.5 = 100.5 > and 1(""(36.5) = 219 + 118.5 < 0, so the profitmaximizing output is Q = 36.5.
°


9.S Taylor Series and the MeanValue Theorem This section considers the socalled "expansion" of a function y=f(x) into what is known as Taylor series (expansion around any point x=Xo). To expand a function y=f(x) around a point Xomeans to transform that function into a polynomial form, in which the coefficients of the various terms are expressed in terms of the derivative values f' (Xo),f" (Xo), etc.  all evaluated at the point of expansion Xo. Taylor's Theorem. Given an arbitrary function cf>(x), f we know the values of f(Xo), f'(Xo), i f"(Xo), etc., then this function can be expanded around the point Xoas follows:
where P" represents the nthdegree polynomial and R,. denotes a remainder which can be denoted by the socalled Lagrange form of the remainder:
R
"

,I,.(D+l)(P)
'f'
(n+I)!
(x XOl+1
with P being some number between x and Xo. Here n! is the "n factorial", defined as n!  n(nl)(n2) ... (3)(2)(I). Remark. When n=O, the Taylor series reduce to the socalled meanvalue theorem: cf>(x) Po + = or
Ro
= cf>(Xo) cf>'(p)(X Xo) + 
96
This meanvalue theorem states that the difference between the value of the function P being some point between x and "0.
q, at "0 and at any other x value can be expressed as the product of the difference (x  "0) and q,'(P) with
Figure 9.8
Remark.
If "0 = 0, then Taylor series reduce to the socalled Maclaurin series:
where P is a point between 0 and x. Example: q,(x) Expand the function
=
111 +x Since q,(I) 114 = 114 = 112 and
around the point "0= 1, with n=4. q,'(x) q,"(x) q,(3)(x)
= =
(1 +X)"2, q,'(I) 6(1
=
+x)"\ q,(3)(1) = 3/8 q,(4)(X) = 24(1 +xrs, q,(4)(1) = 3/4
= 2(1 +X)"3, q,"(1)
97
we obtain the following Taylor series: q,(x) = 112  114 (xI)
+
1/8 (xI)"  1116 (xI)"
+
1/32 (x4)4
+ R".
9.6 NthDerivative Test
A relative extremum of the function f can be equivalently defined as follows: A function f(x) attains a relative maximum (minimum) value at Xoif f(x)  f(Xo) is nonpositive (nonnegative) for values of x in some neighborhood of Xo.
Assume f(x) has finite, continuous derivatives up to the desired order at function can be expanded around x=Xo as a Taylor series:
f(x)  f(xJ fll(XJ  f/(XJ(X  xJ + ~ (x  XJ2 + ...
x= x; then the
+ __ n!
r")(xJ
(xx
0
)"+
r"")(p) (xxJ"" (n+ I)!
From the above expansion, we have the following test: NthDerivative will be (a) (b) (c) A relative maximum if N is an even number and fN)(Xo) < 0. A relative minimum if N is an even number and «N)(Xo) > 0. An inflection point if N is odd. y Test. If f' (Xo) = 0, and if the first nonzero derivative value at Xoencountered
in successive derivation is that of the Nth derivative, «N)(Xo)~ 0, then the stationary value f(Xo)
Example;
24
=
(7x)'.
= 24(77) = 0, and f4)(7) = is a minimum point such that f(7) = 0.
Since f/(7) = 4(77)3 = 0, fll(7) = 12(77)2 = 0, flll(7)
>
0, so x=7
98
x.
DERIVATIVES OF EXPONENTIAL AND LOGARITHMIC
FUNCTIONS
Exponential functions, as well as the closely related logarithmic functions, have important applications in economics, especially in connection with growth problems, and in economic dynamics in general. This chapter gives some basic properties and derivatives of exponential and logarithmic functions. 10.1 The Nature of Exponential Functions In its simple version, the exponential function may be represented in the form:
y = f{t) = b'
(b > 1)
where b denotes a fixed base of the exponent. Its generalized version has the form: y = abet Remark. y = abet = a(bey. Thus we can consider be as a base of the exponent. It changes exponent from ct to t and changes base b to be.
If the base is the irrational number denoted by the symbol e=2.71828 ... , the function:
is referred to the natural exponential function, which can alternatively denote y = a exp{rt). Remark. It can be proved that e may be defined as the limit:
e
=
lim f(n)  lim (1
~c.
~OD
+.!. )D. n
10.2 Logarithmic Functions For the exponential function y=b' and the natural exponential function y=e', taking the log of y to the base b (denoted by log, y) and the base e (denoted by loge y) respectively, we
101
have t and t
=
=
log, Y
logo Y
 In y. The following rules are familiar to us: Rules: (a) (b) (c) (d) (e) In(uv) In(u/v)
= In u + In v = In u  In v
b)
[log of product] [log of quotient] [log of power] [conversion of log base] [inversion of log base]
In ua = a In u log, u = (log, e)(loge u) log, e = lI(lo~
Properties of Log. (a) (b) (c) (d) (e) (t) Remark. In Yl = In Yz In Y1 0 iff y, iffy,
=
> In Yz
Y
Yz > Yz
<
<
1
iff log Y asy asy
<0
Y= 1 log Y'" 00 log y ... oo t
iff log Y = 0
oo
O the respective inverse functions of the exponential
functions y
= log, Y and t = = b' and y = e.
In yare
10.3 Derivatives of Exponential and Logarithmic Functions
(a)
dint (it
1
t
(b)
(c)
d f(,) _e __ dt
f'(t) ef(,)
102
(d)
~ lnf(t) _ f'(t) dt f(t)
Examples; (a) (b) (c) (d) (e) Let y Let y Let y
= a/at = lit y = ln r'. Since y = In t" = c In t, so dy/dt = c(l/t) Let y = fin e. Then dy/dt = 3t2ln t2 + 2f/t = 2t2(l +3
In at, Then dy/dt
= = =
e'. Then dy/dt
e'l. Then dy/dt
= =
re" _e'l
In t)
The Case of Base b (a)
(b)
d logbt dt

1 tlnh
(c)
(d)
d 1 fi(t) _ f'(t) 1 dt ogb f(t) lnh
Proof of (a). Since
b' 
elnb'  ellnb,
then (d/dt)bl
=
(d/dt)elln b
=
(In b)(elln b)
=
b' In b.
Proof of (b), Since Iogbt
=
(Iogbe)(Ioget)
(d/dt)(Iogbt) Examples; (a)
=
=
(l/In b)In t, b)In t]
(d/dt)[(lIin
=
(lIIn b)(lIt)
103
An Application. Example; Find dy/dx from y In y Differentiating
=
xaecx"". Taking the natural log of both sides, we have
=
a In x
+
lex  c.
both with respect to x, we get
Y
Thus
1 dy _ ~ + k
dx
x
dy _ (a/x dx
+
k)y  (a/x
+
k)x
aekxc
104
XI. THE CASE OF MORE THAN ONE CHOICE VARIABLE This chapter develops a way of finding the extreme values of an objective function that involves two or more choice variables. As before, our attention will be focused heavily on relative extrema, and for this reason we should often drop the adjective "relative," with the understanding that, unless otherwise specified, the extrema referred to are relative. 11.1 The Differential Version of Optimation Condition This section shows the possibility of equivalently expressing the derivative version of first and second conditions in terms of differentials. Consider the function z=f(x). dz = f'(x)dx. Since f'(x)=O, which implies dz=O, is the necessary condition for extreme values, so dz=O is also the necessary condition for extreme values. This firstorder condition requires that dz=O as x is varied. In such a context, with dx seO, dz=O if and only if f'(x)=O. What about the sufficient conditions in terms of secondorder differentials? Differentiating dz=f'(x)dx, we have Recall that the differential of z=f(x) is
d2z
iii!
d(dz)  d[f'(x)dx]
 d[f'(x)]dx  f"(X)dx2
Note that the symbols d2z and dx? are fundamentally different. d2z means the secondorder differential of z; but dx2 means the squaring of the firstorder differential dx. Thus, from the above equation, we have d2z < 0 (d2z > 0) if and only if f"(x) < 0 (f"(x) > 0). Therefore, the secondorder sufficient condition for maximum (minimum) of z=f(x) is d2z < 0 (d2z > 0).
111
11.2 Extreme Values of a Function of Two Variables For a function of one variable, an extreme value is represented graphically by the peak of a hill or the bottom of a valley in a twodimensional graph. With two choice variables, the graph of the function z=f(x,y) becomes a surface in a 3space, and while the extreme values are still to be associated with peaks and bottoms.
s
(0)
%
(b)
FirstOrder Condition For the function z=f(x,y), the firstorder necessary condition for an extremum again involves dz=O for arbitrary values of dx and dy: an extremum point must be a stationary point, at which z must be constant for arbitrary infinitesimal changes of two variables x and y. In the present twovariable case, the total differential is
Thus the equivalent derivative version of the firstorder condition dz=O is fx = fy = 0 or ai/ax = af/ay = 0
112
As in the earlier discussion, the firstorder condition is necessary, but not sufficient. To develop a sufficient condition, we must look to the secondorder total, which is related to secondorder partial derivatives. SecondOrder Partial Derivatives From the function z=f(x,y), we can have two firstorder partial derivatives, fx and fy. Since fx and fy are themselves functions of x, we can find secondorder partial derivatives:
f
xx
iiI_fx
ax
a
or
ax2
a2z
;;;;
a ax ax [az] a az] ay ( ay
fyy
IE
~fy
ay
or
a2z ay2
==
flt)'
=
a2z axay a2z ayax
a
a ay ax [az] a ax ay [az]
f
yx iii
==
The last two are called cross (or mixed) partial derivatives. Remark. Even though fxyand fyxhave been separately defined, they willaccording to Young's theorem, be identical with each other, as long as the two cross partial derivatives are both continuous. In fact, this theorem applies also to functions of three or more variables. Given z=g(u,v,w), for instance, the mixed partial derivatives will be characterized by guv = g...., guw = gwu,etc. provided these partial derivatives are continuous. Example: Find all secondorder partial derivatives of z = x3 + 5xy  y2. The first partial
derivatives of this function are fx = 3x2 + 5y and fy = 5x  2y Thus, fxx= 6x, fyx= 5, fxy= 5, and fyy= 2. As expected, fyx= fxy. Example: For z = x2e·y,its first partial derivatives are
113
Thus, the second partial derivatives fxx
are
=
2eY 'YX f
=
Zxe? 'XY f
=
Zxe? 'YY f
=
x2eY
SecondOrder
Total Differentials
From the first total differential
we can obtain the secondorder
d2z=d(d zx+o(dz) d )
total differential d2z:
o(dz) d y __
ox
Y
oy
 ~(fdx
ox
x
+f dy)dx + ~(f
oy
x
dx+ f dy)dy
Y
 [fxxtU +ixydy]tU
+ [fyxtU +iyydy ]dy
 fn dx2 + fxy dydx + fyx dxdy + fyy dy2  f n dx2 + 2f xy dxdy + f yy dy2 [if fxy  fyx]
Example: Given z = x3 + 5xy  y2, find dz and dz2
dz
= =
fxdx (3x2
+ fydy + 5y)dx +
(5x  2y)dy
d2z = fxxdx2 + 2fxydxdy + fyydy2 = 6xdx2 + lOdxdy  2dy2 SecondOrder Condition sufficient condition for:
Using the concept of d~, we can state the secondorder (a) (b) A maximum of z A minimum of z
= =
f(x,y) if d2z f(x,y) if d2z
<
0 for any values of dx and dy, not both zero; 0 for any values of dx and dy, not both zero.
>
114
For operational convenience, secondorder differential conditions can be translated into equivalent conditions on secondorder derivatives. The actual translation would require a knowledge of quadratic forms, which will be discussed in the next section: But we may first introduce the main result here. Main Result: For any values of dx and dy, not both zero, d2z < 0 iff fxx < 0; fyy < 0; and fxxfyy > (fxy)2 d2z > 0 iff fxx > 0; fyy > 0; and fxxfyy > (fxy)2 From the first and secondorder conditions, we obtain conditions for relative extremum: Conditions for Maximum: fx > (fxy)2 Conditions for Minimum: fx > (fxy)2 Example: fx
= =
fy
= =
0 (necessary condition) and fxx < 0; fyy < 0; and fxxfyy
fy
0 (necessary condition) and fxx > 0; fyy > 0; and fxxfyy
Find the extreme values of z
=
8x3 + 2xy  3x2 + y2 + 1. 2x + 2y fxy = 2 fy
fxx = 48x  6 , Setting fx
=
24x2 + 2y  6x ,
fyy = 2
=
=
0 and fy
=
0, we have
24x2 + 2y  6x = 0 2y + 2x = 0 Then y = x and thus from 24x2 + 2y  6x, we have 24x2  8x = 0 which yields two solutions for x: XI = 0 and XI = 113. Since fxx(O,O) = 6 and fyy(O,O) = 2, it is impossible fxxfyy ~ (fxy)2 = 4, so the point
(XI;YI) = (0,0) is not extreme point. For the solution (X2'YI) = (113, 113), we find that fxx = 10 > 0, fyy = fxy = 2 > 0, and fxxfyy  (fxy)2 = 20  4 > 0, so (x,y,z) = (113, 113, 23/27)
is a relative minimum point. Example: z = x + 2ey  e"  eZy. Letting fx = 1  eX = 0 and fy = 2e  2eZy 0, we have X = o and y= 112. Since fxx = eX, fyy = 4eZy, and fxy = 0, we have fxx(O, 112) = 1 < 0,
115
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.