UNIT 5

UNCONSTRAINED
OPTIMISATION

Structure
5.0 Objectives
5.1 Introduction
5.2 Unconstrained Optimisation
5.2.1 First Order Condition
5.2.2 Second Order Condition

5.3 Some Economic Problems

5.4 Let Us Sum Up
5.5 Key Words
5.6 Some Useful Books
5.7 Answers or Hints to Check Your Progress

5.0

OBJECTIVES

After going through this unit we should able to:

solve simple unconstrained optimisation problems; and

solve simple economic problems dealing with unconstrained optimisation.

5.1

INTRODUCTION

In this unit, our objective is to find and explain the necessary and sufficient
conditions for any unconstrained optimisation problem. Optimisation in
mathematical sense (and therefore for our purpose) means either maximisation
or, minimisation of different mathematical functions. Optimisation is broadly
distinguishable in two respects viz., Unconstrained Optimisation (which is our
current focus) and Constrained Optimisation (which we discuss in the next
unit). Unconstrained optimisation deals with those problems whose domain is
not compressed by any constrained. We are going to use simple calculus to
solve these kinds of optimisation problems. First, let us we find and discuss
the necessary first order conditions for unconstrained optimisation before
moving on to the sufficient condition or, the second order condition. In the
next section, we will focus on some simple economic applications that deal
with unconstrained optimisation.

5.2

UNCONSTRAINED OPTIMISATION

Under optimisation exercise we either maximise or, minimise some

mathematical functions consists of either one variable or, several variables. To
optimize a mathematical function, we need to find necessary first order
conditions from which the optimum values of the variables can be found. But
to find whether the value is maximum or, minimum we have to check the
sufficient second order condition.

5.2.1 First Order Condition

Consider the function y = f ( x) , which can be plotted as shown below. The
function has the maximum point A and the minimum point B. Note that at
these two points the value of y must be stationary. We can say that it is a
22

necessary conditions for an extremum of y that dy = 0 instantaneously as x

varies. This condition is known as the differential version of the first order
condition for an extremum and we have already discussed it in the preceding
unit. As we have said earlier, this condition is necessary but not sufficient for
either a maximum or, a minimum. Therefore, it is clear that the necessary first
order condition for both maxima and minima are the same.

Unconstrained
Optimisation

Fig 5.1: First order condition for optimum

We will see that the above condition is equivalent to the derivative version of
dy
the first order condition
= 0 or, f ' ( x) = 0 . We note that when there is no
dx
change in x (dx = 0), dy will automatically be zero. But this, of course, is not
what the first order condition is all about. What the first order condition
requires is that dy be zero as x is varied, that is, as arbitrary (positive or,
negative, but not zero) infinitesimal changes of x occur. In such a context,
with dx 0 , dy can be zero if and only if f ' ( x) = 0 . Thus, the derivative
condition f ' ( x) = 0 and the differential condition dy = 0 for arbitrary nonzero
values of dx are indeed equivalent.
Now suppose y = f ( x1 , x2 ,..., xn) then in this case, the first order condition for
y
optimisation is given by the n-equation viz.,
= f 1 ( x) = 0 ,
x1

y
y
= f n ( x) = 0 . From these n equations, we can find
= f 2 ( x) = 0 , ,
xn
x 2
x1* ,x2* ,...,xn * .

Example: Find the optimal vales of y = x3 x in the domain < x < + .

dy
= 0 , which can be
dx
transformed to 3 x 2 1 = 0 - (I). From equation (I) we get at optimum either
1
1
x* = +
or, x* =
. Putting these values of x, we get optimum y i.e., y* .
3
3

Solution: The necessary first order condition gives

Example: Find the optimal values of z = y 2 xy + x 2 on the domain

< x < + and < y < + .
23

Extreme Values
and Optimisation

z
z
= 0 and
= 0 , from
x
y
which we can get y + 2 x = 0 and 2 y x = 0 . We solve this and get x* = 0 and
y* = 0 . Finally putting these optimum values into the original equation we get
z* = 0 .
Solution: The necessary first order condition gives

5.2.2 Second Order Condition

For the time being, let us assume the function z = f ( x, y ) can give rise two
z
z
first order partial derivatives, fx and fy
. Since fx is itself a function
x
y
of x (as well as y), we can measure the rate of change of fx with respect to x,
while y remains fixed, by a particular second order (or, second) partial
2 z
derivative denoted by either fxx or, 2 .
x

2 z z
. The notation fxx has a double subscript
( fx ) Or,
x 2 x x
x
signifying that the primitive function f has been differentiated partially with
2 z
d2z
except for
respect to x twice, whereas the notation 2 resembles that of
x
dx 2
the use of the partial symbol. In a perfectly analogues manner, we can use the

2 z z
second partial derivative fyy ( fy ) or, 2 to denote the rate of
y
y
y y
change of fy with respect to y, while x is held constant.
fxx

Recall however, that fx is also a function of x. Hence, there can be written two
2 z
z
2 z
z
.
and fyx
yx y x
xy x y
These are called cross (or, mixed) partial derivatives because each measures
the rate of change of one first order partial derivative with respect to the
other variable. We can also note that using total differential equation we get
(z )
(z )
d 2 z d (dz )
dx +
dy .
x
y

more partial derivatives:

fxy

Using the concept of d 2 z , we can state the second order sufficient condition
for a maximum of z = f ( x, y ) as follows: d 2 z < 0 for arbitrary values of dx
and dy, not both zero (a). The rationale behind is very similar to that of the
d 2 z condition for the one variable case and it can be explained by means of
Fig 5.2.1, which depicts the birds-eye view of a surface. Let point A on the
surface the point lying directly above the point (x0, y0) in the domain
satisfy the first order condition. Then the point A is a prospective candidate
for a maximum. Whether it in fact qualifies depends on the surface
configuration in the neighborhood of A. If an infinitesimal movement away
from A in any direction along the surface invariable results in a decrease in z
that is, if dz < 0 for arbitrary values of dx and dy, not both zero A is a pick
of a dome. Given that dz = 0 at point A, however, the condition dz < 0 at point
in the neighborhood of A amounts to the stipulation that dz is decreasing, that
is, d 2 z d (dz ) < 0 , for arbitrary values of dx and dy, not both zero. Thus (a)
constitutes a sufficient condition for identifying a stationary value as a
24

maximum of z. Analogous reasoning would show that a counterpart second

order sufficient condition for identifying value as a minimum of z = f(x, y) is
as follows: d 2 z > 0 for arbitrary values of dx and dy, not both zero (b).

Unconstrained
Optimisation

The reason why (a) and (b) are sufficient conditions but not necessary
conditions, is that it is again possible for d2z to take a zero value at a
maximum or, a minimum. For this reason, second order necessary conditions
must be stated with weak inequalities as follows: For maximum of z:
d 2 z 0 for arbitrary values of dx and dy, not both zero. For minimum of z:
d 2 z 0 for arbitrary values of dx and dy, not both zero (c). In the following,
however, we pay more attention to the second order sufficient conditions.
For operational convenience, second order differential conditions can be
translated into equivalent conditions on second order derivatives. In the twovariable case, this would entail restrictions on the signs of the second partial
derivatives fxx, fyy and fxy . The actual translation would require knowledge of
quadratic forms. But we may first introduce the main results here: For any
values of dx and dy, not both zero, we have

d 2 z < 0 if and only if fxx < 0 , fyy < 0 , and fxxfyy > fxy 2
d 2 z > 0 if and only if fxx > 0 , fyy > 0 , and fxxfyy > fxy 2 .

Note that the sign of d2z hinges not only on fxx and fyy , which have to do with
the surface configuration around point A, but also on the cross partial
derivative fxy . The role played by this latter partial derivative is to ensure that
the surface in question will yield (two-dimensional) cross sections with the
same type of configuration in all possible directions.
Table 5. 1: Conditions for Relative Extremum: z=f(x, y)
Condition

Maximum

Minimum

First order necessary condition

fx = fy = 0

fx = fy = 0

Second order sufficient condition

fxx, fyy < 0

fxx, fyy > 0

and fxxfyy > fxy

and fxxfyy > fxy

The above result, together with the first order condition, enables us to
construct Table 5.1. It should be understood that all the second partial
derivatives therein are to be evaluated at the stationary point where fx = fy = 0.
It should also be stressed that the second order sufficient condition is not
necessary for an extremum. In particular, if a stationary value is characterised
by fxxfyy = fxy 2 in violation of that condition, that stationary value may
nevertheless turn out to be an extremum. On the other hand, in the case of
another type of violation, with a stationary point characterised by fxxfyy < fxy 2 ,
we can identify that point as a saddle point, because the sign of d2z will in that
case be indefinite (positive for same values of dx and dy, but negative for
others).
For n-variable case we can generalise the above second order condition. The
satisfaction of the first order condition earmarks certain values of z as the
stationary values of the objective function. If at a stationary value of z we find
that d2z is positive definite, this will suffice to establish that value of z as a
minimum. Analogously, the negative definiteness of d2z is a sufficient
condition for the stationary value to be maximum.
25

Extreme Values
and Optimisation

When there are n choice variables, the objective function may be expressed as
z = f ( x1 , x2 ,..., xn ) .
The total differential will then be dz = f1dx1 + f 2 dx2 + ... + f n dxn so that the
necessary condition for extremum (dz = 0 for arbitrary dxi) means that all the
n first order partial derivatives are required to be zero.
The second order differential d2z will again be a quadratic form, which can be
expressed as an n n array. The coefficients of that array, properly arranged,
will now give the (symmetric) Hessian
f11
f
| H |= 21
L

f n1

f12
f 22
L
fn2

L
L
L
L

f1n
f 2 n
L

f nn

f12
f
with principle minors | H1 |= f11 , | H 2 |= 11
,, | H n |=| H | . The
f 21 f 22
second order sufficient condition for extremum is, as before, that all the n
principal minors be positive (for a minimum in z) and that they duly alternate
in sign (for a maximum in z), the first one being negative.
In summary, then if we concentrate on the determinantal test we have the
criteria as listed in Table 5.2, which is valid for an objective function of any
number of choice variables. As special cases, we can have n = 1 or, n = 2.
When n = 1, the objective function is z = f(x), and the condition for
maximization, f1 = 0 and | H1 |< 0 , reduce to f ' ( x) = 0 and f " ( x) < 0 , exactly
as we learned above. Similarly, when n = 2, the objective function is
z = f ( x1 , x2 ) , so that the first order condition for maximum is f1 = f 2 = 0 ,
whereas the second order sufficient condition becomes f11 < 0 and
f11 f12
= f11 f 22 f12 2 > 0 , which is merely a restatement of the
f

21 f 22
information given above.
Table 5. 2: Determinantal Test for Relative Optimum: z = f ( x1 , x2 ,..., xn )

Condition
First order necessary condition

Maximum

f1 = f 2 = ... = f n

Second order sufficient condition (1)i | H i |> 0 i =1,,n

Minimum

f1 = f 2 = ... = f n
| H i |> 0 i=1,,n

Example: Find the extreme value(s) of z = x 2 + xy + 2 y 2 + 3 and determine

whether they are maximum or, minimum.
Solution: The relevant first order conditions are the following:

fx = 2 x + y = 0 and fy = x + 4 y = 0 . Solving these two equations we get x* = 0

and y* = 0. Note that fxx = 2 > 0 and fyy = 4 > 0 and as fxy = 1 we have
fxxfyy > fxy 2 . So at x* = 0 and y* = 0 the function z is minimized.
Example: Find the extreme value(s) of z = x 2 + xy y 2 + 2 x + y and
determine whether they are maximum or, minimum.

26

Solution: The relevant first order conditions are the following:

fx = 2 x + y + 2 = 0 and fy = x 2 y + 1 = 0 . Solving these two equations we

get x* = 5/3 and y* = 4/3. Now note that fxx = 2 < 0 and fyy = 2 < 0 and as
fxy = 1 we have fxxfyy > fxy 2 . So at x* = 5/3 and y* = 4/3 the function z is
maximised.

Unconstrained
Optimisation

Check Your Progress 1

1)

Find the extreme value(s) of z = x 2 + xy y 2 + x + 5 y and determine

whether they are maximum or, minimum.

2)

Find the extreme value(s) of y = x 2 6 and determine whether they are

maximum or, minimum.

3)

Find the extreme value(s) of y = x3 2 x 2 + x 6 and determine whether

they are maximum or, minimum.

27

Extreme Values
and Optimisation

4)

Find the extreme value(s) of z = x 2 2 x y 2 and determine whether

they are maximum or, minimum.

5.3

SOME ECONOMIC PROBLEMS

In this section, we will give some economic application concerning

unconstrained optimisation. Our first example is concerning the multi-product
firm who takes the prices as given i.e., she produces under competitive
structure. The second example is also regarding multi-product firm but here
the firm acts as a monopolist where she can choose how much to produce and
at what price. In both the cases the firms objective is to maximise her own
profit where profit is defined as net revenue minus net cost. These objectives
we take for simplicity i.e., one can think of other objective of the firm viz.,
revenue maximisation, but that is not going to be discussed in this unit. For
simplicity again, we take cost function to be the same in both example.
Example 1: This example is regarding the problem of a multi-product firm.
Let us first postulate a two-product firm under circumstances of pure
competition. Since with pure competition the prices of both commodities must
be taken as exogenous, these will be given by p1 = 5 and p2 = 3, respectively.
Accordingly, the firms revenue function will be R = p1q1 + p2q2 = 5q1 + 3q2,
where qi represents the output level of the ith product per unit of time. The
firms cost function is assumed to be C = 2q12 + 2q22 + q1q2.

C
= 4q1 + q2 (the marginal cost of the first product) is a function
q1
not only of q1 but also of q2. Similarly, the marginal cost of the second
product also depends, in part, on the output level of the first product. Thus,
according to the assumed cost function, the two commodities are seen to be
technically related in production.
Note that

The profit function of this hypothetical firm can now be written readily as
= R C = 5q1 + 3q2 2q12 2q2 2 q1q2 , a function of two choice variables
(q1 and q2) and two price parameters. It is our task to find the levels of q1 and
q2, which in combination, will maximise . For this purpose, we find the first
order partial derivatives of the profit function:

= 5 4q1 q2 and 2
= 3 4q2 q1 - (a)
q1
q2

28

Setting these both equal to zero, to satisfy the necessary condition for
maximum, we get the two simultaneous equations 4q1 + q2 = 5 and
4 p p2 17
q1 + 4q2 = 3 , which yield the unique solution q1* = 1
= and
15
15
4
p

p
7
1
q2* = 2
= . Thus, the maximum profit that the firm will get is
15
15
*
= 5.51 (approx).

Unconstrained
Optimisation

To be sure that this does represent a maximum profit, let us check the second
order condition. The second partial derivatives, obtainable by partial
differentiation of equation (a), give us the following Hessian:

12 4 1

| H |= 11
=

21 22 1 4
Since | H1 |= 4 < 0 and | H 2 |= 15 > 0 , the Hessian matrix (or, d2z) is negative
definite, and the solution does maximise the profit. In fact, since the signs of
the principal minors do not depends on where they are evaluated, d2z is in this
case everywhere negative definite. Thus, the objective function must be
strictly concave, and the maximum profit found above is actually a unique
absolute maximum.
Example 2: This example again regarding a multi-product firm but with the
modification that in earlier case the firm face given price i.e., multi-product
firm under competitive structure, in this example the multi-product firm is a
monopoly in both the product i.e., the firm can decide how much quantity to
be produced and at the same time she can choose any price for her product.
Now by virtue of this new market-structure assumption, the revenue function
must be modified to reflect the fact that the prices of the two products will
now vary with their output levels (which are assumed to be identical with their
sales levels, no inventory accumulation being contemplated in the model). The
exact manner in which prices will vary with output levels is, of course, to be
found in the demand functions for the firms two products.

Suppose that the demand facing the monopolistic firm, are as follows:
q1 = 40 2 p1 + p2 and q2 = 15 + p1 p2 - (a)
These equations reveal that the two commodities are related in consumption;
specifically, they are substitute goods, because an increase in the price of one
will raise the demand for the other. As given, equation (a) expresses the
quantities demanded q1 and q2 as functions of prices, but for our present
purposes, it will be more convenient to have prices p1 and p2 expressed in
terms of the sales volumes q1 and q2, that is, to have average revenue
functions for the two products. Since equation (a) can be rewritten as
2 p1 + p2 = q1 40 and p1 p2 = q2 15 , we may (considering q1 and q2 as
parameters) apply Cramers rule to solve for p1 and p2 as follows:
p1 = 55 q1 q2 and p2 = 70 q1 2q2 - (b)
These constitute the desired average-revenue functions, since p1 AR1 and
p2 AR2 . Consequently, the firms total-revenue function can be written as:
29

R = p1q1 + p2 q2

Extreme Values
and Optimisation

= (55 q1 q2 )q1 + (70 q1 2q2 )q2

= 55q1 + 70q2 2q1q2 q12 2q2 2
If we again assume the total cost function to be C = q12 + q1q2 + q2 2 , then the
profit function will be:

= R C
= 55q1 + 70q2 3q1q2 2q12 3q2 2 - (c),
which is an objective function with two choice variables. Once the profit
maximizing output levels q1* and q2* are found, however, the optimal prices
p1* and p2* are easy enough to find from equation (b).
The objective function yields the following first and second partial
derivatives:

1 = 55 3q1 4q2

11 = 4

2 = 70 3q1 6q2

12 = 21 = 3

22 = 6

To satisfy the first order condition for a maximum of , we must have

1 = 2 = 0 ; that is 4q1 + 3q2 = 55 and 3q1 + 6q2 = 70 , thus the solution
2
output levels (per unit of time) are (q1*, q2*) = (8, 7 ).
3
Upon substitution of this result into equation (b) and (c), respectively, we find
1
2
1
that p1* = 39 p2* = 46
and
(per unit of time)
* = 488
3
3
3

4 3
3 6 , we have | H1 |= 4 < 0

*
and | H 2 |= 15 > 0 , so that the value of does represent the maximum profit.
Here, the signs of the principle minors are again independent of where they
are evaluated. Thus, the Hessian matrix is everywhere negative definite,
implying that the objective function is strictly concave and that it has a unique
absolute maximum.

Inasmuch

5.4

as

the

Hessian

is

LET US SUM UP

In this unit we emphasise an important tool, which has enormous application

in economic viz., unconstrained optimisation. Unconstrained optimisation
problem deals with optimisation (i.e., either minimisation or, maximisation)
problems where the domain of objective function is not limited by any
constrained. There are many real life economic problems for which
unconstrained optimisation technique is necessary. Any problem regarding
profit maximisation, we use the technique of unconstrained optimisation. In
this unit, we emphasise the technique, how to solve unconstrained
optimisation problems. For that we first introduced the first order necessary
conditions for optimisation. We saw that if there are n choice variables we
must have n number of equality as first order condition to solve the optimum
value of each variable. We also saw that the first order conditions are same for
both maximisation and minimisation problem, and therefore we cant get
whether by putting the optimum values of variables we get maximum or,

30

minimum value of the objective function. To solve this kind of problem we

therefore introduce second order sufficient condition for optimisation. We saw
second order condition is different for maximisation and minimisation
exercise. We also saw that for a given objective function and for given
optimum values of choice variables, if the corresponding Hessian matrix is
negative definite at optimum then the optimum values give the maximum
value of the objective function. Analogously, if the Hessian matrix is positive
definite at optimum then the optimum values give the minimum value of the
objective function. Therefore, from the first order condition (supported by
second order condition), we can determine the maximum or, the minimum
value of the objective function. There is a notion of necessary second order
condition which tells that the necessary second order condition for
maximization is that, the Hessian matrix should be negative semi-definite and
the necessary second order condition for minimization is that, the Hessian
matrix should be positive semi-definite.

Unconstrained
Optimisation

Finally, we give some interesting economic applications that clear the

techniques of unconstrained optimisation problem. For that our first example
is a problem regarding multi-product firm under competitive structure. Then
we give an example again concerning multi-product firm but now the firm
acting as monopolist in all markets.

5.5

KEY WORDS

Positive Definite Matrix: A symmetric matrix A is positive definite if it is

true that for every vector x, the real number which results from the
multiplication x'Ax > 0 for all nonzero x. Such a matrix has real positive
eigenvalues.
Positive Semi-definite Matrix: A symmetric matrix A is positive definite if it
is true that for every vector x, the real number which results from the
multiplication x'Ax 0 for all nonzero x.
Negative Definite Matrix: A symmetric matrix A is negative definite if it is
true that for every vector x, the real number which results from the
multiplication x'Ax 0 for all nonzero x. Such a matrix has real negative
eigenvalues.
Negative Semi-definite Matrix: A symmetric matrix A is negative semidefinite if it is true that for every vector x, the real number which results from
the multiplication x'Ax 0 for all nonzero x.
Stationary Point: In calculus, a stationary point is referred to a point on a
graph where the tangent to the graph is parallel to the x-axis or, equivalently,
where the derivative of the function equals zero.
Necessary and Sufficient Condition for Optimum: Given a function
z = f ( x1 , x2 ,..., xn ) , the conditions are obtained as the following:
Condition
First order necessary condition

Maximum

f1 = f 2 = ... = f n

Second order sufficient condition (1) | H i |> 0

i

Minimum

f1 = f 2 = ... = f n

i =1,,n | H i |> 0 i=1,,n

31

Extreme Values
and Optimisation

5.6

SOME USEFUL BOOKS

Henderson, James M. and Quandt, Richard E. (1980), Microeconomic Theory,

McGrow-Hill Book Company, New York.
Allen, R.G.D. (1938), Mathematical Analysis for Economists, St. Martin's
Press, New York.
Varian, Hal (1992), Microeconomic Analysis, W.W. Norton & Company, Inc.,
Yew York.

5.7

ANSWERS OR HINTS TO CHECK YOUR

PROGRESS

Check Your Progress 1

32

1)

See the example given in section 5.2.2.

2)

The objective function is y = x 2 6 . Therefore, from the first order

condition for optimisation, we get 2 x = 0 . Solving this equation we have
x = 0. The second order condition gives that d 2 y = 2 > 0 , which is not a
function of x. Therefore, we have d2y = 2 at x = 0. So at x = 0, we reach
the minima of the objective function.

3)

The objective function is y = x3 2 x 2 + x 6 . Therefore, we get the first

1
order condition 3 x 2 4 x + 1 = 0 or, either x = or, x = 1. Note that
3
1
d 2 y = 6 x 4 . So, d2y > 0 at x = 1 and d2y < 0 at x = . Therefore, at x
3
1
= 1, we have the minima and at x = .
3

4)

The objective function is z = x 2 2 x y 2 . The necessary first order

condition yields 2 x 2 = 0 and 2y = 0. From these two first order
conditions we get x = 1 and y = 0. The corresponding Hessian matrix is
2 0
| H |=
, so we get that | H1 |= 2 > 0 and | H 2 |= 4 > 0 . Therefore,
0 2
we have the minimum value of z at x =1 and y = 0.