Constrained Optimization1

EC 115 Lecture Notes

Marjorie S. Muyrong, M.A.2

In most cases in economics, the optimization problem involves a constraint. Hence, if the
objective function we are maximizing is 𝑧 = 𝑓(𝑥, 𝑦) as in Figure 5.1, then we have a
constrained maximum. In other words, the essence of optimization is to find the values of the
choice variables that will yield that desired extremum of your objective function given that
there is another function that puts forth a limit to what level of utility a consumer can achieve.
Graphically, we have Figure 5.1:

Figure 5.1 Constrained Optimization

Source: Chiang and Wainwright (2005)

1
Lectures notes are in part adopted from Chiang and Wainwright (2005) and Dumlao (2016).
2
Department of Economics, Ateneo de Manila University, mmuyrong@[Link]

1
For instance, the utility function 𝑈 = 𝑈(𝑥1 , 𝑥2 ) = 𝑥1 𝑥2 + 2𝑥1 tells us that utility increases
infinitely as consumption of two goods increases. In other words, the principle of non-satiation
may be invoked—if money was not a problem, then 𝑥1 ∗ = ∞ and 𝑥2 ∗ = ∞. If there is no budget
constraint to consider, consumers will buy infinite amounts of various goods and there is no
utility maximization problem to solve.

A. Utility Maximization

The Constrained Optimization Process involves (1) identifying the objective function and its
constraint function, and then (2) choosing the values of the choice variables that will provide
the desired level of the dependent variable given the constraint. Hence if the objective function
is the utility function, 𝑈 = 𝑈(𝑥1 , 𝑥2 ), what must be its constraint? The constraint must be his
budget such that his expenditures given the prices of good 1 and good 2 cannot go above his
budget, i.e. 𝑀 = 𝑔(𝑥1 , 𝑥2 ) where 𝑀 is the budget and 𝑔(𝑥1 , 𝑥2 ) is the expenditures function.
What must be the choice variables? The consumer must therefore choose the level of
consumption for both goods that he can afford and would maximize his utility. Graphically, we
have Figure 5.2:

Figure 5.2 Utility Maximization Problem

𝑥2

A
B C
𝑥2 ∗ U3
U2
U1
M
𝑥1
𝑥1 ∗
The consumer in Figure 5.2 can afford only market bundles A and B as both lie in the budget
constraint. Market bundle C on the other hand is not affordable. The consumer would therefore
choose market bundle B since it gives him the highest level of utility given his budget. Going
back to the utility function 𝑈 = 𝑈(𝑥1 , 𝑥2 ) = 𝑥1 𝑥2 + 2𝑥1 , let us say that the budget constraint

2
faced by the firm is 60 = 𝑃1 𝑥1 + 𝑃2 𝑥2 = 4𝑥1 + 2𝑥2 . This means that the budget of the
consumer is only Php 60 and the prices of good 1 and good 2 are Php 4 and Php 2 respectively.
To solve the constrained utility maximization problem, we now have to transform the budget
constraint (using the multiplier 𝜆) into utility and attach it to the utility function. Therefore, we
are able to set up a modified utility function which is called the Lagrangean function:

ℒ = 𝑈(𝑥1 , 𝑥2 ) + 𝜆(𝑀 − 𝑔(𝑥1 , 𝑥2 ))

(6.1)
ℒ = 𝑥1 𝑥2 + 2𝑥1 + 𝜆(60 − 4𝑥1 − 2𝑥2 )

𝜕ℒ
As (6.1) shows, = 𝜆, which means that the multiplier is the marginal utility from money
𝜕𝑀
𝜕ℒ
you have. On the other hand, 𝜕𝑔 = −𝜆, which means that you decrease your utility by 𝜆, every

time you spend Php 1. However, that spending is actually transformed into goods which of
course will again allow the consumer to obtain utility. This, in other words, means that setting
up the Lagrangean function for the utility maximization problem is the case of how a consumer
transforms his utility from holding money into utility of consuming goods.

Given that the utility function is now given by the Lagrangean function, the first-order
conditions (FOCs) are the result of setting 𝑑ℒ = ℒ1 𝑑𝑥1 + ℒ2 𝑑𝑥2 + ℒ𝜆 𝑑𝜆 = 0. Hence, we have
the following FOCs:

𝜕ℒ
ℒ1 = = 𝑥2 + 2 − 4𝜆 = 0
𝜕𝑥1

𝜕ℒ
ℒ2 = = 𝑥1 − 2𝜆 = 0 (6.2)
𝜕𝑥2

𝜕ℒ
ℒ𝜆 = = 60 − 4𝑥1 − 2𝑥2 = 0
𝜕𝜆

The FOCs may therefore be transformed into its 𝐴𝑥 = 𝑑 format to find the solutions:

−2 1 −4
| 0 0 −2|
−128
𝑥1 ∗ = 60 2 0 = =8 (6.3)
0 1 −4 −16
|1 0 −2|
4 2 0

3
 0 −2 −4
|1 0 −2|
𝑥2 ∗ = 4 60 0 = −224 = 14 (6.4)
0 1 −4 −16
|1 0 −2 |
4 2 0
0 1 −2
|1 0 0|
𝜆∗ = 4 2 60 = −64 = 4 (6.5)
0 1 −4 −16
|1 0 −2|
4 2 0

If the consumer chooses market bundle B to maximize his utility given his budget, this means
than he consumes 8 units of good 1 and 14 units of good 2. At the same time, his utility
maximization problem tells us that he also obtains a marginal utility of 4 utils from every peso
of his budget.
The second-order condition (SOC) for a constrained maximum is now going to be a little
different compared to the unconstrained problem. The SOC is defined by:

𝑑2 ℒ = ℒ11 𝑑𝑥12 + 2ℒ12 𝑑𝑥1 𝑑𝑥2 + ℒ22 𝑑𝑥22 < 0 (6.6)

Note too that our expenditure function, which is part of our constraint, must have no change
since the budget is already set at:

𝑑𝑀 = 𝑑𝑔(𝑥1 , 𝑥2 )
0 = 𝑔1 𝑑𝑥1 + 𝑔2 𝑑𝑥2
(6.7)
𝑔1
− 𝑑𝑥1 = 𝑑𝑥2
𝑔2
Inserting (6.7) and rearranging the SOC:

𝑔1 𝑔1 2
𝑑 2 ℒ = ℒ11 𝑑𝑥12 + 2ℒ12 𝑑𝑥1 (− 𝑑𝑥1 ) + ℒ22 (− 𝑑𝑥1 ) < 0
𝑔2 𝑔2
𝑔1 𝑔1 2
𝑑2 ℒ = ℒ11 𝑑𝑥12 − 2 ℒ12 𝑑𝑥12 + ℒ22 ( ) 𝑑𝑥12 < 0
𝑔2 𝑔2
(6.8)
2
𝑑𝑥12 2
𝑔1 𝑔22
𝑑 ℒ = 2 [ℒ11 𝑔2 − 2 ℒ + ℒ22 (𝑔1 )2 ] < 0
𝑔2 𝑔2 12
𝑑𝑥12
𝑑 ℒ = 2 [ℒ11 𝑔22 − 2ℒ12 𝑔1 𝑔2 + ℒ22 𝑔12 ] < 0
2
𝑔2

4
We can then solve for the Hessian matrix for this modified utility function:
ℒ11 ℒ12 ℒ1𝜆 ℒ11 ℒ12 −𝑔1
̅
|𝐻 | = |ℒ21 ℒ22 ℒ2𝜆 | = | ℒ21 ℒ22 −𝑔1 |
ℒ𝜆1 ℒ𝜆2 ℒ𝜆𝜆 −𝑔1 −𝑔2 0
(6.9)
= (0 + ℒ12 𝑔1 𝑔2 + ℒ21 𝑔1 𝑔2 ) − (ℒ11 𝑔22 +0+ ℒ22 𝑔12 )
= −ℒ11 𝑔22 + 2ℒ12 𝑔1 𝑔2 − ℒ22 𝑔12

Such Hessian matrix is defined to be bordered since the second partial derivatives and the cross
partial derivatives are bordered by the negative of the first partial derivatives from the budget
̅ | in (6.9) is actually opposite signs with the terms in brackets in equation
constraint. Note that |𝐻
(6.8). We already know that to ensure a maximum, we need 𝑑 2 ℒ < 0. To ensure this, we
̅ | > 0.
require |𝐻
To ensure a maximum in the original problem, we now set up the bordered Hessian matrix and
get its determinant:
ℒ11 ℒ12 ℒ1𝜆 0 1 −4
̅ ℒ
|𝐻 | = | 21 ℒ22 ℒ2𝜆 | = | 1 0 −2| = (0 + 8 + 8) − (0) = 16 > 0 (6.10)
ℒ𝜆1 ℒ𝜆2 ℒ𝜆𝜆 −4 −2 0

̅ | > 0, then we have a maximum. (What’s the proof of |𝐻

Because |𝐻 ̅ | > 0 for a maximum?)

B. Cost Minimization

For the case of the firm, their goal, apart from maximizing profit, is minimizing cost. If you
look at it, in the unconstrained profit maximization problem, a firm is able to determine its
desirable level of production. Simultaneous to this therefore is to choose the combination of
input that will minimize his costs given a desired quantity. Hence if the objective function is
the cost function 𝐶 = 𝐶(𝐾, 𝐿) = 𝑟𝐾 + 𝑤𝐿 , what must be its constraint? The constraint must
be his budget such that his production function 𝑄0 = 𝑓(𝐾, 𝐿) = 𝐾 𝛼 𝐿1−𝛼 . What must be the
choice variables? The consumer must therefore choose the level of capital and labor that will
give the firm the set production level while minimizing cost. Graphically, we have Figure 5.3:

5
 Figure 5.3 Cost Minimization Problem

A
B
K* C
Q0
Q-1

C1 C2

L* L

The firm in Figure 5.3 is choosing only between input combinations B and C as both lie in the
desired level of production 𝑄0 . Input combination A on the other hand would lead to a lower
level of production. The consumer would therefore choose input combination B since it gives
the firm the lowest level of cost given its set production level. To solve the constrained cost
minimization problem, we now have to transform the production constraint (using the
multiplier 𝜆) into costs and attach it to the cost function. Therefore, we are able to set up a
modified cost function which is also the Lagrangean function for this scenario:

ℒ = 𝐶(𝐾, 𝐿) + 𝜆(𝑄0 − 𝑓(𝐾, 𝐿) )

(6.7)
𝛼 1−𝛼 )
ℒ = 𝑟𝐾 + 𝑤𝐿 + 𝜆(𝑄0 − 𝐾 𝐿

𝜕ℒ
As (6.1) shows, 𝜕𝑄 = 𝜆, which means that the multiplier is the marginal cost of producing an
0

𝜕ℒ
additional unit of good. On the other hand, 𝜕𝑓 = −𝜆, which means that you must gain revenues

by 𝜆 by producing a good. However, producing goods of course translates into costs. This, in
other words, means that setting up the Lagrangean function for the cost minimization problem
is the case of how a firm transforms production of goods into costs.

6
The first-order conditions (FOCs) are given by the following:

𝜕ℒ
ℒ𝐾 = = 𝑟 − 𝜆𝛼𝐾 𝛼−1 𝐿1−𝛼 = 0
𝜕𝐾
𝜕ℒ
ℒ𝐿 = = 𝑤 − 𝜆(1 − 𝛼)𝐾 𝛼 𝐿−𝛼 = 0 (6.8)
𝜕𝐿
𝜕ℒ
ℒ𝜆 = = 𝑄0 − 𝐾 𝛼 𝐿1−𝛼 = 0
𝜕𝜆

The FOCs this time cannot be transformed into matrix algebra since they are not linear
equations. The first and second FOCs are thus rearranged and equated together and solve for
either capital or labor in terms of the other:
𝑟 𝑤
𝜆= =
𝛼𝐾 𝛼−1 𝐿1−𝛼 (1 − 𝛼)𝐾 𝛼 𝐿−𝛼

(1 − 𝛼)𝐾 𝛼 𝐿−𝛼 𝑤
=
𝛼𝐾 𝛼−1 𝐿1−𝛼 𝑟
𝐾 𝛼 𝐿−𝛼 𝑤 𝛼
=
𝐾 𝛼−1 𝐿1−𝛼 𝑟 (1 − 𝛼)

𝐾 𝛼−𝛼+1 𝑤 𝛼
=
𝐿1−𝛼+𝛼 𝑟 (1 − 𝛼)

𝐾 𝑤 𝛼
=
𝐿 𝑟 (1 − 𝛼)
𝛼 𝑤
𝐾= 𝐿
(1 − 𝛼) 𝑟

Then, we bind the resulting equation to the constraint:

𝛼 𝑤 𝛼 1−𝛼
𝑄0 = [ 𝐿] 𝐿
(1 − 𝛼) 𝑟

𝛼 𝑤 𝛼 1−𝛼+𝛼
𝑄0 = [ ] 𝐿
(1 − 𝛼) 𝑟

𝛼 𝑤 𝛼
𝑄0 = [ ] 𝐿
(1 − 𝛼) 𝑟

𝑄0
𝐿=
𝛼 𝑤 𝛼
[ ]
(1 − 𝛼) 𝑟

7
 ∗
(1 − 𝛼) 𝑟 𝛼
𝐿 = 𝑄0 [ ]
𝛼 𝑤

To get the equilibrium level of capital, we use 𝐿∗ to the result of combining the first two FOCs:
𝛼 𝑤 ∗
𝐾= 𝐿
(1 − 𝛼) 𝑟

𝛼 𝑤 (1 − 𝛼) 𝑟 𝛼
𝐾= 𝑄 [ ]
(1 − 𝛼) 𝑟 0 𝛼 𝑤

(1 − 𝛼) 𝑟 −1 (1 − 𝛼) 𝑟 𝛼
𝐾 = 𝑄0 [ ] [ ]
𝛼 𝑤 𝛼 𝑤

∗
(1 − 𝛼) 𝑟 𝛼−1
𝐾 = 𝑄0 [ ]
𝛼 𝑤

We can also get 𝜆∗ :

𝑟
𝜆=
𝛼𝐾 ∗ 𝛼−1 𝐿∗1−𝛼
𝑟
𝜆=
𝛼−1 𝛼−1 1−𝛼
(1 − 𝛼) 𝑟 (1 − 𝛼) 𝑟 𝛼
𝛼 {𝑄0 [ 𝛼 𝑤 ] } {𝑄0 [ 𝛼 𝑤 ] }

𝑟
𝜆=
(1 − 𝛼) 𝑟 1−𝛼
𝛼{ 𝛼 𝑤}
1 𝛼 𝑤 1−𝛼
𝜆=𝑟 { }
𝛼 (1 − 𝛼) 𝑟
𝑟𝛼 𝛼 1−𝛼
𝜆= { 𝑤}
𝛼 (1 − 𝛼)
1−𝛼
𝑟𝛼 1
𝜆 = 𝛼{ 𝑤}
𝛼 (1 − 𝛼)
𝑟 𝛼 𝑤 1−𝛼
𝜆= 𝛼
𝛼 (1 − 𝛼)1−𝛼
𝑟 𝛼 𝑤 1−𝛼
𝜆∗ = ( ) ( )
𝛼 1−𝛼

If the firm chooses input combination B to minimize its cost given its profit-maximizing
quantity, this means than its rents 𝐾 ∗ units of capital and hires 𝐿∗ number of laborers. How do

8
we interpret 𝜆∗ ? To ensure a minimum, we now set up the bordered Hessian matrix and get its
determinant:
ℒ𝐾𝐾 ℒ𝐾𝐿 ℒ𝐾𝜆
̅ | = | ℒ𝐿𝐾
|𝐻 ℒ𝐿𝐿 ℒ2𝐿𝜆 |
ℒ𝜆𝐾 ℒ𝜆𝐿 ℒ𝜆𝜆
−(𝛼 − 1)𝜆𝛼𝐾 𝛼−2 𝐿1−𝛼 −(1 − 𝛼)𝜆𝛼𝐾 𝛼−1 𝐿−𝛼 −𝛼𝐾 𝛼−1 𝐿1−𝛼
̅ | = | −𝛼𝜆(1 − 𝛼)𝐾 𝛼−1 𝐿−𝛼
|𝐻 −(−𝛼)𝜆(1 − 𝛼)𝐾 𝛼 𝐿−𝛼−1 −(1 − 𝛼)𝐾 𝛼 𝐿−𝛼 |
−𝛼𝐾 𝛼−1 𝐿1−𝛼 −(1 − 𝛼)𝐾 𝛼 𝐿−𝛼 0

(+) (−) (−)

̅ | = |(−) (+) (−)| = [0 + (−) + (−)] − [(+) + 0 + (+)]
|𝐻
(−) (−) 0
̅ | = [−] − [+] = [−] < 0
|𝐻

̅ | < 0, then we have a minimum. (What’s the proof of |𝐻

Because |𝐻 ̅ | < 0 for a minimum?)