CHAPTER 9

Constrained Optimisation
Rational economic agents are assumed to make choices that maximise
their utility or prot. But their choices are usually constrained
for example the consumers choice of consumption bundle is constrained
by his income. In this chapter we look at methods for solving optimisa-
tion problems with constraints: in particular the method of Lagrange
multipliers. We apply them to consumer choice, cost minimisa-
tion, and other economic problems.

1. Consumer Choice
Suppose there are two goods available, and a consumer has preferences represented by the
utility function u(x
1
, x
2
) where x
1
and x
2
are the amounts of the goods consumed. The prices
of the goods are p
1
and p
2
, and the consumer has a xed income m. He wants to choose his
consumption bundle (x
1
, x
2
) to maximise his utility, subject to the constraint that the total
cost of the bundle does not exceed his income. Provided that the utility function is strictly
increasing in x
1
and x
2
, we know that he will want to use all his income.
The consumers optimisation problem is:
max
x
1
,x
2
u(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m
The objective function is u(x
1
, x
2
)
The choice variables are x
1
and x
2
The constraint is p
1
x
1
+p
2
x
2
= m
There are three methods for solving this type of problem.
1.1. Method 1: Draw a Diagram and Think About the Economics
x
1
x
2
P
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Provided the utility function is well-behaved (increasing
in x
1
and x
2
with strictly convex indierence curves), the
highest utility is obtained at the point P where the budget
constraint is tangent to an indierence curve.
The slope of the budget constraint is:
p
1
p
2
and the slope of the indierence curve is: MRS =
MU
1
MU
2
(See Chapters 2 and 7 for these results.)
153
154 9. CONSTRAINED OPTIMISATION
Hence we can nd the point P:
(1) It is on the budget constraint:
p
1
x
1
+p
2
x
2
= m
(2) where the slope of the budget constraint equals the slope of the indierence curve:
p
1
p
2
=
MU
1
MU
2
In general, this gives us two equations that we can solve to nd x
1
and x
2
.
Examples 1.1: A consumer has utility function u(x
1
, x
2
) = x
1
x
2
. The prices of the two
goods are p
1
= 3 and p
2
= 2, and her income is m = 24. How much of each good will she
consume?
(i) The consumers problem is:
max
x
1
,x
2
x
1
x
2
subject to 3x
1
+ 2x
2
= 24
The utility function u(x
1
, x
2
) = x
1
x
2
is Cobb-Douglas (see Chapter 7). Hence it is
well-behaved. The marginal utilities are:
MU
1
=
u
x
1
= x
2
and MU
2
=
u
x
2
= x
1
(ii) The optimal bundle:
(1) is on the budget constraint:
3x
1
+ 2x
2
= 24
(2) where the budget constraint is tangent to an indierence curve:
p
1
p
2
=
MU
1
MU
2

3
2
=
x
2
x
1
(iii) From the tangency condition: 3x
1
= 2x
2
. Substituting this into the budget constraint:
2x
2
+ 2x
2
= 24
x
2
= 6
and hence x
1
= 4
The consumers optimal choice is 4 units of good 1 and 6 units of good 2.
1.2. Method 2: Use the Constraint to Substitute for one of the Variables
If you did A-level maths, this may seem to be the obvious way to solve this type of problem.
However it often gives messy equations and is rarely used in economic problems because it
doesnt give much economic insight. Consider the example above:
max
x
1
,x
2
x
1
x
2
subject to 3x
1
+ 2x
2
= 24
From the constraint, x
2
= 12
3x
1
2
. By substituting this into the objective function, we can
write the problem as:
max
x
1

12x
1

3x
2
1
2

9. CONSTRAINED OPTIMISATION 155

So, we have transformed it into an unconstrained optimisation problem in one variable. The
rst-order condition is:
12 3x
1
= 0 x
1
= 4
Substituting back into the equation for x
2
we nd, as before, that x
2
= 6. It can easily be
checked that the second-order condition is satised.
1.3. The Most General Method: The Method of Lagrange Multipliers
This method is important because it can be used for a wide range of constrained optimisation
problems. For the consumer problem:
max
x
1
,x
2
u(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m
(1) Write down the Lagrangian function:
L(x
1
, x
2
, ) = u(x
1
, x
2
) (p
1
x
1
+p
2
x
2
m)
is a new variable, which we introduce to help solve the problem.
It is called a Lagrange Multiplier.
(2) Write down the rst-order conditions:
L
x
1
= 0
L
x
2
= 0
L

= 0
(3) Solve the rst-order conditions to nd x
1
and x
2
.
Provided that the utility function is well-behaved (increasing in x
1
and x
2
with strictly con-
vex indierence curves) then the values of x
1
and x
2
that you obtain by this procedure will
be the optimum bundle.
1
Examples 1.2: A consumer has utility function u(x
1
, x
2
) = x
1
x
2
. The prices of the two
goods are p
1
= 3 and p
2
= 2, and her income is m = 24. Use the Lagrangian method to nd
how much of each good she will she consume.
(i) The consumers problem is, as before:
max
x
1
,x
2
x
1
x
2
subject to 3x
1
+ 2x
2
= 24
Again, since the utility function is Cobb-Douglas, it is well-behaved.
(ii) The Lagrangian function is:
L(x
1
, x
2
, ) = x
1
x
2
(3x
1
+ 2x
2
24)
1
In this Workbook we will simply use the Lagrangian method, without explaining why it works. For some
explanation, see Anthony & Biggs section 21.2.
156 9. CONSTRAINED OPTIMISATION
(iii) First-order conditions:
L
x
1
= x
2
3 x
2
= 3
L
x
2
= x
1
2 x
1
= 2
L

= (3x
1
+ 2x
2
24) 3x
1
+ 2x
2
= 24
(iv) Eliminating from the rst two equations gives:
3x
1
= 2x
2
Substituting for x
1
in the third equation:
2x
2
+ 2x
2
= 24 x
2
= 6 and hence x
1
= 4
Thus we nd (again) that the optimal bundle is 4 units of good 1 and 6 of good 2.
1.4. Some Useful Tricks
1.4.1. Solving the Lagrangian rst-order conditions. In the example above, the rst two
equations of the rst-order conditions are:
x
2
= 3
x
1
= 2
There are several (easy) ways to eliminate to obtain 3x
1
= 2x
2
. But one way, which is
particularly useful in Lagrangian problems when the equations are more complicated, is to
divide the two equations, so that cancels out:
x
2
x
1
=
3
2

x
2
x
1
=
3
2
Whenever you have two equations:
A = B
C = D
where A, B, C and D are non-zero algebraic expressions, you can write:
A
C
=
B
D
.
1.4.2. Transforming the objective function. Remember from Chapter 7 that if preferences
are represented by a utility function u(x
1
, x
2
), and f is an increasing function, then f(u)
represents the same preference ordering. Sometimes we can use this to reduce the algebra
needed to solve a problem. Suppose we want to solve:
max
x
1
,x
2
x
3
4
1
x
1
4
2
subject to 3x
1
+ 2x
2
= 24
Preferences are represented by a Cobb-Douglas utility function u = x
3
4
1
x
1
4
2
, so they could
equivalently be represented by:
lnu =
3
4
lnx
1
+
1
4
lnx
2
or 4 lnu = 3 lnx
1
+ lnx
2
You can check that solving the problem:
max
x
1
,x
2
3 lnx
1
+ lnx
2
subject to 3x
1
+ 2x
2
= 24
gives exactly the same answer as the original problem, but that the equations are simpler.
9. CONSTRAINED OPTIMISATION 157
1.5. Well-Behaved Utility Functions
We have seen that for the optimisation methods described above to work, it is important
that the utility function is well-behaved: increasing, with strictly convex indierence curves.
Two examples of well-behaved utility functions are:
The Cobb-Douglas Utility Function
u(x
1
, x
2
) = x
a
1
x
b
2
The Logarithmic Utility Function
u(x
1
, x
2
) = a lnx
1
+b lnx
2
(We know from Chapter 7 that the Cobb-Douglas function is well-behaved. The logarithmic
function is a monotonic increasing transformation of the Cobb-Douglas, so it represents the
same preferences and is also well-behaved.) Two other useful forms of utility function are:
The CES Utility Function
2
u(x
1
, x
2
) =

ax

1
+bx

where a and b are positive parameters, and is a parameter greater than 1.

The Quasi-Linear Utility Function
u(x
1
, x
2
) = v(x
1
) +x
2
where v(x
1
) is any increasing concave function.
Both of these are well-behaved. To prove it, you can show that:
- the function is increasing, by showing that the marginal utilities are positive
- the indierence curves are convex as we did for the Cobb-Douglas case (Chapter
7 section 4.2) by showing that the MRS gets less negative as you move along an
indierence curve.
Exercises 9.1: Consumer Choice
(1) (1) Use Method 1 to nd the optimum consumption bundle for a consumer with utility
function u(x, y) = x
2
y and income m = 30, when the prices of the goods are p
x
= 4
and p
y
= 5. Check that you get the same answers by the Lagrangian method.
(2) A consumer has a weekly income of 26 (after paying for essentials), which she
spends on restaurant Meals and Books. Her utility function is u(M, B) = 3M
1
2
+B,
and the prices are p
M
= 6 and p
B
= 4. Use the Lagrangian method to nd her
optimal consumption bundle.
(3) Use the trick of transforming the objective function (section 1.4.2) to solve:
max
x
1
,x
2
x
3
4
1
x
1
4
2
subject to 3x
1
+ 2x
2
= 24
Further Reading and Exercises
Varian Intermediate Microeconomics, Chapter 5, covers the economic theory of
Consumer Choice. The Appendix to Chapter 5 explains the same three methods for
solving choice problems that we have used in this section.
Jacques 5.5 and 5.6
Anthony & Biggs 21.2, 22.2, and 22.3
2
Constant Elasticity of Substitution this refers to another property of the function
158 9. CONSTRAINED OPTIMISATION

2. Cost Minimisation
Consider a competitive rm with production function F(K, L). Suppose that the wage rate
is w, and the rental rate for capital is r. Suppose that the rm wants to produce a particular
amount of output y
0
at minimum cost. How much labour and capital should it employ?
The rms optimisation problem is:
min
K,L
(rK +wL) subject to F(K, L) = y
0
The objective function is rK +wL
The choice variables are K and L
The constraint is F(K, L) = y
0
We can use the same three methods here as for the consumer choice problem, but we will
ignore method 2 because it is generally less useful.
2.1. Method 1: Draw a Diagram and Think About the Economics
L
K
P
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
F(K, L) = y
0
Draw the isoquant of the production function representing
combinations of K and L that can be used to produce
output y
0
. (See Chapter 7.)
The slope of the isoquant is: MRTS =
MPL
MPK
Draw the isocost lines where rK +wL = constant.
The slope of the isocost lines is:
w
r
Provided the isoquant is convex, the lowest cost is
achieved at the point P where an isocost line is tangent to
the isoquant.
Hence we can nd the point P:
(1) It is on the isoquant:
F(K, L) = y
0
(2) where the slope of the isocost lines equals the slope of the isoquant:
w
r
=
MPL
MPK
This gives us two equations that we can solve to nd K and L.
Examples 2.1: If the production function is F(K, L) = K
1
3
L
2
3
, the wage rate is 5, and the
rental rate of capital is 20, what is the minimum cost of producing 40 units of output?
(i) The problem is:
min
K,L
(20K + 5L) subject to K
1
3
L
2
3
= 40
9. CONSTRAINED OPTIMISATION 159
The production function F(K, L) = K
1
3
L
2
3
is Cobb-Douglas, so the isoquant is convex.
The marginal products are:
MPK =
F
K
=
1
3
K

2
3
L
2
3
and MPL =
F
L
=
2
3
K
1
3
L

1
3
(ii) The optimal choice of inputs:
(1) is on the isoquant:
K
1
3
L
2
3
= 40
(2) where an isocost line is tangent to it:
w
r
=
MPL
MPK

5
20
=
2
3
K
1
3
L

1
3
1
3
K

2
3
L
2
3

1
8
=
K
L
(iii) From the tangency condition: L = 8K. Substituting this into the isoquant:
K
1
3
(8K)
2
3
= 40
K = 10
and hence L = 80
(iv) With this optimal choice of K and L, the cost is: 20K+5L = 2010+580 = 600.
2.2. The Method of Lagrange Multipliers
For the cost minimisation problem, the method is analagous to the one for consumer choice:
min
K,L
(rK +wL) subject to F(K, L) = y
0
(1) Write down the Lagrangian function:
L(K, L, ) = rK +wL (F(K, L) y
0
)
(2) Write down the rst-order conditions:
L
K
= 0,
L
L
= 0,
L

= 0
(3) Solve the rst-order conditions to nd K and L.
Provided that the isoquants are convex, this procedure obtains the optimal values of K and L.
Examples 2.2: If the production function is F(K, L) = K
1
3
L
2
3
, the wage rate is 5, and the
rental rate of capital is 20, use the Lagrangian method to nd the minimum cost of producing
40 units of output.
(i) The problem is, as before:
min
K,L
(20K + 5L) subject to K
1
3
L
2
3
= 40
and since the production function is Cobb-Douglas, the isoquant is convex.
(ii) The Lagrangian function is:
L(K, L, ) = 20K + 5L (K
1
3
L
2
3
40)
160 9. CONSTRAINED OPTIMISATION
(iii) First-order conditions:
L
K
= 20
1
3
K

2
3
L
2
3
60 = K

2
3
L
2
3
L
L
= 5
2
3
K
1
3
L

1
3
15 = 2K
1
3
L

1
3
L

= (K
1
3
L
2
3
40) K
1
3
L
2
3
= 40
(iv) Dividing the rst two equations to eliminate :
60
15
=
K

2
3
L
2
3
2K
1
3
L

1
3
which simplies to L = 8K
Substituting for L in the third equation:
K
1
3
(8K)
2
3
= 40 K = 10 and hence L = 80
(v) Thus, as before, the optimal choice is 10 units of capital and 80 of labour, which means
that the cost is 600.
Exercises 9.2: Cost Minimisation
(1) (1) A rm has production function F(K, L) = 5K
0.4
L. The wage rate is w = 10 and
the rental rate of capital is r = 12. Use Method 1 to determine how much labour
and capital the rm should employ if it wants to produce 300 units of output.
What is the total cost of doing so?
Check that you get the same answer by the Lagrangian method.
(2) A rm has production function F(K, L) = 30(K
1
+L
1
)
1
. Use the Lagrangian
method to nd how much labour and capital it should employ to produce 70 units
of output, if the wage rate is 8 and the rental rate of capital is 2.
Note: the production function is CES, so has convex isoquants.
Further Reading and Exercises
Jacques 5.5 and 5.6
Anthony & Biggs 21.1, 21.2 and 21.3
Varian Intermediate Microeconomics, Chapter 20, covers the economics of Cost
Minimisation.
9. CONSTRAINED OPTIMISATION 161
_

3. The Method of Lagrange

Multipliers
To try to solve any problem of the form:
max
x
1
,x
2
F(x
1
, x
2
) subject to g(x
1
, x
2
) = c
or:
min
x
1
,x
2
F(x
1
, x
2
) subject to g(x
1
, x
2
) = c
you can write down the Lagrangian:
L(x
1
, x
2
, ) = F(x
1
, x
2
) (g(x
1
, x
2
) c)
and look for a solution of the three rst-order conditions:
L
x
1
= 0,
L
x
2
= 0,
L

= 0
Note that the condition
L

= 0 always gives you the constraint.

3.1. Max or Min?
But if you nd a solution to the Lagrangian rst-order conditions, how do you know whether
it is a maximum or a minimum? And what if there are several solutions? The answer is that
in general you need to look at the second-order conditions, but unfortunately the SOCs for
constrained optimisation problems are complicated to write down, so will not be covered in
this Workbook.
However, when we apply the method to economic problems, we can often manage without
second-order conditions. Instead, we think about the shapes of the objective function and
the constraint, and nd that either it is the type of problem in which the objective function
can only have a maximum, or that it is a problem that can only have a minimum. In such
cases, the method of Lagrange multipliers will give the required solution.
If you look back at the examples in the previous two sections, you can see that the Lagrangian
method gives you just the same equations as you get when you draw a diagram and think
about the economics what the method does is to nd tangency points.
But when you have a problem in which either the objective function or constraint doesnt have
the standard economic properties of convex indierence curves or isoquants you cannot rely
on the Lagrangian method, because a tangency point that it nds may not be the required
maximum or minimum.
Examples 3.1: Standard and non-standard problems
(i) If the utility function is well-behaved (increasing, with convex indierence curves) the
problem:
max
x
1
,x
2
u(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m
can be solved by the Lagrangian method.
(ii) If the production function has convex isoquants the problem:
min
K,L
(rK +wL) subject to F(K, L) = y
0
can be solved by the Lagrangian method.
162 9. CONSTRAINED OPTIMISATION
(iii) Consider a consumer choice problem with concave indierence curves:
max
x
1
,x
2
u(x
1
, x
2
) subject to 5x
1
+ 4x
2
= 20
where u(x
1
, x
2
) = x
2
1
+x
2
2
.
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
x
1
x
2
5
4
P
The tangency point P is the point on
the budget constraint where utility is
minimised.
Assuming that the amounts x
1
and x
2
cannot be negative, the point on the bud-
get constraint where utility is maximised
is x
1
= 0, x
2
= 5.
If we write the Lagrangian:
L(x
1
, x
2
, ) = x
2
1
+x
2
2
(5x
1
+4x
2
20)
the rst-order conditions are:
2x
1
= 5, 2x
2
= 4, 5x
1
+ 4x
2
= 20
Solving, we obtain x
1
= 2.44, x
2
= 1.95. The Lagrangian method has found the
tangency point P, which is where utility is minimised.
3.2. The Interpretation of the Lagrange Multiplier
In the examples of the Lagrangian method so far, we did not bother to calculate the value
of satisfying the rst-order conditions we simply eliminated it and solved for the choice
variables. However, this value does have a meaning.
In an optimisation problem with objective function F(x, y)
and constraint g(x, y) = c, let F

be the value of the

objective function at the optimum. The value of the
Lagrange multipler indicates how much F

would
increase if there were a small increase in c:
=
dF

dc
This means, for example, that for the cost minimisation problem:
min
K,L
(rK +wL) subject to F(K, L) = y
the Lagrange multiplier tells us how much the cost would increase if there were a small in-
crease in the amount of output to be produced that is, the marginal cost of output.
We will prove this result using dierentials (see Chapter 7). The Lagrangian is:
L = rK +wL (F(K, L) y)
and the rst-order conditions are:
r = F
K
w = F
L
F(K, L) = y
9. CONSTRAINED OPTIMISATION 163
These equations can be solved to nd the optimum values of capital K

and labour L

, and
the Lagrange multiplier

. The cost is then C

= rK

+wL

.
K

, L

and C

all depend on the the level of output to be produced, y. Suppose there is

small change in this amount, dy. This will lead to small changes in the optimal factor choices
and the cost, dK

, dL

, and dC

. We can work out how big the change in cost will be:
First, taking the dierential of the constraint F(K

, L

) = y:
dy = F
K
dK

+F
L
dL

then, taking the dierential of the cost C

= rK

+wL

and using the rst-order conditions:

dC

= rdK

+wdL

F
K
dK

F
L
dL

dy

dC

dy
=

So the value of the Lagrange multiplier tells us the marginal cost of output.
Similarly for utility maximisation:
max
x
1
,x
2
u(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m
the Lagrange multiplier gives the value of
du

dm
, which is the marginal utility of income.
This, and the general result in the box above, can be proved in the same way.
Examples 3.2: The Lagrange Multiplier
In Examples 2.2, for the production function F(K, L) = K
1
3
L
2
3
, with wage rate 5 and
rental rate of capital 20, we found the minimum cost of producing 40 units of output. The
Lagrangian and rst-order conditions were:
L(K, L, ) = 20K + 5L (K
1
3
L
2
3
40)
60 = K

2
3
L
2
3
15 = 2K
1
3
L

1
3
K
1
3
L
2
3
= 40
Solving these we found the optimal choice K = 10, L = 80, with total cost 600.
Substituting these values of K and L back into the rst rst-order condition:
60 = 10

2
3
80
2
3
= 15
Hence the rms marginal cost is 15. So we can say that producing 41 units of output
would cost approximately 615. (In fact, in this example, marginal cost is constant: the
value of does not depend on the number of units of output.)
164 9. CONSTRAINED OPTIMISATION
3.3. Problems with More Variables and Constraints
The Lagrangian method can be generalised in an obvious way to solve problems in which
there are more variables and several constraints. For example, to solve the problem:
max
x
1
,x
2
,x
3
,x
4
F(x
1
, x
2
, x
3
, x
4
) subject to g
1
(x
1
, x
2
, x
3
, x
4
) = c
1
and g
2
(x
1
, x
2
, x
3
, x
4
) = c
2
the Lagrangian would be:
L(x
1
, x
2
, x
3
, x
4
,
1
,
2
) = F(x
1
, x
2
, x
3
, x
4
)
1
(g
1
(x
1
, x
2
, x
3
, x
4
)c
1
)
2
(g
2
(x
1
, x
2
, x
3
, x
4
)c
2
)
giving six rst-order conditions to solve for the optimal choice.
Exercises 9.3: The Method of Lagrange Multipliers
(1) (1) Find the Lagrange Multiplier and hence the marginal cost of the rm in Exercises
9.2, Question 2.
(2) Find the optimum consumption bundle for a consumer with utility u = x
1
x
2
x
3
and income 36, when the prices of the goods are p
1
= 1, p
2
= 6, p
3
= 10.
(3) A consumer with utility function u(x
1
, x
2
) has income m = 12, and the prices
of the goods are p
1
= 3 and p
2
= 2. For each of the following cases, decide
whether the utility function is well-behaved, and determine the optimal choices:
(a) u = x
1
+x
2
(b) u = 3x
2/3
1
+x
2
(c) u = min(x
1
, x
2
)
(4) A rm has production function F(K, L) =
1
4

K
1/2
+L
1/2

. The wage rate is

w = 1 and the rental rate of capital is r = 3.
(a) How much capital and labour should the rm employ to produce y units of
output?
(b) Hence nd the cost of producing y units of output (the rms cost function).
(c) Dierentiate the cost function to nd the marginal cost, and verify that it
is equal to the value of the Lagrange multiplier.
(5) By following similar steps to those we used for the cost minimisation problem in
section 3.2 prove that for the utility maximisation problem:
max
x
1
,x
2
u(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m
the Lagrange multiplier is equal to the marginal utility of income.
Further Reading and Exercises
Anthony & Biggs Chapter 22 gives a general formulation of the Lagrangian method,
going beyond what we have covered here.
9. CONSTRAINED OPTIMISATION 165
_
`

4. Some More Examples of

Constrained Optimisation
Problems in Economics
4.1. Production Possibilities
Robinson Crusoe spends his 8 hour working day either shing or looking for coconuts. If he
spends t hours shing and s hours looking for coconuts he will catch f(t) = t
1
2
sh and nd
c(s) = 3s
1
2
coconuts. Crusoes utility function is u = lnf + lnc.
The optimal pattern of production in this economy is the point on the production possibility
frontier (ppf) where Crusoes utility is maximised.
The time taken to catch f sh is t = f
2
, and the time take to nd c coconuts is s =

c
3

2
.
Since he has 8 hours in total, the ppf is given by:
f
2
+

c
3

2
= 8
To nd the optimal point on the ppf we have to solve the problem:
max
c,f
(lnc + lnf) subject to f
2
+

c
3

2
= 8
Note that the ppf is concave, and the utility function is well-behaved (it is the log of a
Cobb-Douglas). Hence the optimum is a tangency point, which we could nd either by the
Lagrangian method, or using the condition that the marginal rate of substitution must be
equal to the marginal rate of transformation see the similar problem on Worksheet 7.
4.2. Consumption and Saving
A consumer lives for two periods (work and retirement). His income is 100 in the rst period,
and zero in the second. The interest rate is 5%. His lifetime utility is given by:
U(c
1
, c
2
) = c
1
2
1
+ 0.9c
1
2
2
If he consumes c
1
in the rst period he will save (100 c
1
). So when he is retired he can
consume:
c
2
= 1.05(100 c
1
)
This is his lifetime budget constraint. Rearranging, it can be written:
c
1
+
c
2
1.05
= 100
Hence the consumers optimisation problem is:
max
c
1
,c
2

c
1
2
1
+ 0.9c
1
2
2

subject to c
1
+
c
2
1.05
= 100
which can be solved using the Lagrangian:
L(c
1
, c
2
, ) = c
1
2
1
+ 0.9c
1
2
2

c
1
+
c
2
1.05
100

to determine the optimal consumption and saving plan for the consumers lifetime.
166 9. CONSTRAINED OPTIMISATION
4.3. Labour Supply
A consumer has utility
U(C, R) = 3 lnC + lnR
where C is her amount of consumption, and R is the number of hours of leisure (relaxation)
she takes each day. The hourly wage rate is w = 4 (measured in units of consumption). She
has a non-labour income m = 8 (consumption units). She needs 10 hours per day for eating
and sleeping; in the remainder she can work or take leisure.
Suppose that we want to know her labour supply how many hours she chooses to work.
First we need to know the budget constraint. If she takes R units of leisure, she will work
for 14 R hours, and hence earn 4(14 R). Then she will be able to consume:
C = 4(14 R) + 8
units of consumption. This is the budget constraint. Rearranging, we can write it as:
C + 4R = 64
So we need to solve the problem:
max
C,R
(3 lnC + lnR) subject to C + 4R = 64
to nd the optimal choice of leisure R, and hence the number of hours of work.
Exercises 9.4: More Constrained Optimisation Problems
Complete the solution of each of three problems discussed in this section.
9. CONSTRAINED OPTIMISATION 167

5. Determining Demand Functions

5.1. Consumer Demand
In the consumer choice problems in section 1 we determined the optimal consumption bundle,
given the utility function and particular values for the prices of the goods, and income. If we
solve the same problem for general values p
1
, p
2
, and m, we can determine the consumers
demands for the goods as a function of prices and income:
x
1
= x
1
(p
1
, p
2
, m) and x
2
= x
2
(p
1
, p
2
, m)
Examples 5.1: A consumer with a xed income m has utility function u(x
1
, x
2
) = x
2
1
x
3
2
.
If the prices of the goods are p
1
and p
2
, nd the consumers demand functions for the two
goods.
The optimisation problem is:
max
x
1
,x
2
x
2
1
x
3
2
subject to p
1
x
1
+p
2
x
2
= m
The Lagrangian is:
L(x
1
, x
2
, ) = x
2
1
x
3
2
(p
1
x
1
+p
2
x
2
m)
Dierentiating to obtain the rst-order conditions:
2x
1
x
3
2
= p
1
3x
2
1
x
2
2
= p
2
p
1
x
1
+p
2
x
2
= m
Dividing the rst two equations:
2x
2
3x
1
=
p
1
p
2
x
2
=
3p
1
x
1
2p
2
Substituting into the budget constraint we nd:
p
1
x
1
+
3p
1
x
1
2
= m
x
1
=
2m
5p
1
and hence x
2
=
3m
5p
2
These are the consumers demand functions. Demand for each good is an increasing
function of income and a decreasing function of the price of the good.
5.2. Cobb-Douglas Utility
Note that in Examples 5.1, the consumer spends
2
5
of his income on good 1 and
3
5
on good 2:
p
1
x
1
=
2
5
m and p
2
x
2
=
3
5
m
This is an example of a general result for Cobb-Douglas Utility functions:
A consumer with Cobb-Douglas utility: u(x
1
, x
2
) = x
a
1
x
b
2
spends a constant fraction of income on each good:
p
1
x
1
=
a
a+b
m and p
2
x
2
=
b
a+b
m
168 9. CONSTRAINED OPTIMISATION
You could prove this by re-doing example 5.1 using the more general utility function u(x
1
, x
2
) =
x
a
1
x
b
2
.
This result means that with Cobb-Douglas utility a consumers demand for one good does
not depend on the price of the other good.
5.3. Factor Demands
Similarly, if we solve the rms cost minimisation problem:
min
K,L
(rK +wL) subject to F(K, L) = y
0
for general values of the factor prices r and w, and the output y
0
, we can nd the rms
Conditional Factor Demands (see Varian Chapter 20) that is, its demand for each factor
as a function of output and factor prices:
K = K(y
0
, r, w) and L = L(y
0
, r, w)
From these we can obtain the rms cost function:
C(y, r, w) = rK(y, r, w) +wL(y, r, w)
Exercises 9.5: Determining Demand Functions
(1) (1) A consumer has an income of y, which she spends on Meals and Books. Her utility
function is u(M, B) = 3M
1
2
+B, and the prices are p
M
and p
B
. Use the Lagrangian
method to nd her demand functions for the two goods.
(2) Find the conditional factor demand functions for a rm with production function
F(K, L) = KL. If the wage rate and the rental rate for capital are both equal to
4, what is the rms cost function C(y)?
(3) Prove that a consumer with Cobb-Douglas utility spends a constant fraction of
income on each good.
Further Reading and Exercises
Anthony & Biggs 21.3 and 22.4
Varian Intermediate Microeconomics, Chapters 5 and 6 (including Appendices)
for consumer demand, and Chapter 20 (including Appendix) for conditional factor
demands and cost functions.
9. CONSTRAINED OPTIMISATION 169
Solutions to Exercises in Chapter 9
Exercises 9.1:
(1) MU
x
= 2xy, MU
y
= x
2
Tangency Condition:
2xy
x
2
=
4
5
y =
2
5
x
Budget Constraint: 4x + 5y = 30
6x = 30 x = 5, y = 2.
(2) max
M,B

3M
1
2
+B

s.t. 6M + 4B = 26
L =

3M
1
2
+B

(6M + 4B 26)

3
2
M

1
2
= 6, 1 = 4, 6M + 4B = 26
Eliminate M = 1
Then b.c. 6 + 4B = 26 B = 5.
(3) L = (3 lnx
1
+ lnx
2
) (3x
1
+ 2x
2
24)

3
x
1
= 3,
1
x
2
= 2, 3x
1
+ 2x
2
= 24
Solving x
2
= 3, x
1
= 6.
Exercises 9.2:
(1) min
K,L
(12K + 10L) s.t. 5K
0.4
L = 300
MPL = 5K
0.4
, MPK = 2K
0.6
L
Tangency:
10
12
=
5K
0.4
2K
0.6
L
L = 3K
Isoquant: 5K
0.4
L = 300 15K
1.4
= 300
K = (20)
1
1.4
= 8.50, L = 25.50
Cost=12K + 10L = 357
(2) min
K,L
(8L + 2K) s.t.
30
L
1
+K
1
= 70
L = (8L + 2K)

3
L
1
+K
1
7

8 =
3
L
2
(L
1
+K
1
)
2
, 2 =
3
K
2
(L
1
+K
1
)
2
Eliminate K = 2L. Substitute in:
3
L
1
+K
1
= 7 L = 3.5, K = 7
Exercises 9.3:
(1) From 2
nd
FOC in previous question:
3 = 2K
2
(L
1
+K
1
)
2
.
Substituting L = 3.5, K = 7 = 0.6
(2) max
x
1
,x
2
,x
3
x
1
x
2
x
3
s.t. x
1
+ 6x
2
+ 10x
3
= 36
L = x
1
x
2
x
3
(x
1
+ 6x
2
+ 10x
3
36)
(i) x
2
x
3
= , (ii) x
1
x
3
= 6,
(iii) x
1
x
2
= 10, (iv) x
1
+6x
2
+10x
3
= 36
Solving: (i) and (iii) x
1
= 10x
3
, and
(ii) and (iii) x
2
=
5
3
x
3
.
Substituting in (iv): 30x
3
= 36
x
3
= 1.2, x
1
= 12 and x
2
= 2.
(3) (a) No. Perfect substitutes: u is linear,
not strictly convex. Consumer buys
good 2 only as it is cheaper: x
1
= 0,
x
2
= 6.
(b) Yes. Quasi-linear; x
2/3
1
is concave.
L = (6x
2/3
1
+x
2
) (3x
1
+2x
2
12)
2x
1/3
1
= 3, 1 = 2,
and 3x
1
+ 2x
2
= 12
Eliminating 2x
1/3
1
=
3
2
x
1/3
1
=
4
3
x
1
= 2.37
and from budget constraint:
7.11 + 2x
2
= 12 x
2
= 2.45
(c) No. Perfect complements.
Optimum is on the budget constraint
where x
1
= x
2
:
3x
1
+ 2x
2
= 12 and x
1
= x
2
x
1
= x
2
= 2.4
(4) (a) L = 3K+L

1
4

L
1/2
+K
1/2

24 = K
1/2
, 8 = L
1/2
and
L
1/2
+K
1/2
= 4y. Solving:
K = y
2
, L = 9y
2
.
(b) C = 3K +L = 12y
2
(c) C

= 24y.
From f.o.c. = 24K
1/2
= 24y
(5) L = u(x
1
, x
2
) (p
1
x
1
+p
2
x
2
m)
f.o.c.s: u
1
=p
1
, u
2
=p
2
, p
1
x
1
+p
2
x
2
=m
Constraint: dm = p
1
dx
1
+p
2
dx
2
Utility: du = u
1
dx
1
+ u
2
dx
2
= p
1
dx
1
+
p
2
dx
2
= dm
du
dm
=
Exercises 9.4:
(1) L = (lnc + lnf)

f
2
+

c
3

2
8

1
f
= 2f,
1
c
=
2c
9
, f
2
+

c
3

2
= 8
Solving f = 2, c = 6.
(2) 0.5c

1
2
1
=, 0.45c

1
2
2
=

1.05
, c
1
+
c
2
1.05
= 100
Elim. c
2
=.893c
1
so c
1
+
.893
1.05
c
1
=100
c
1
= 54.04, c
2
= 48.26
(3)
3
C
= ,
1
R
= 4, C + 4R = 64. Solving:
C = 12R, R = 4, C = 48, 10 hrs work.
Exercises 9.5:
(1) L = (3M
1/2
+B)(p
M
M+p
B
By)
3
2
M

1
2
= p
M
, 1 = p
B
, p
M
M+p
B
B = y
Solving for the demand functions:
M = 2.25p
2
B
/p
2
M
, B = y/p
B
2.25p
B
/p
M
170 9. CONSTRAINED OPTIMISATION
(2) L = (rK +wL) (KLy) w = K,
r = L and KL = y. Solving:
L =

ry
w
, K =

wy
r
, C(y, 4, 4) = 8

y
(3) L = x
a
1
x
b
2
(p
1
x
1
+p
2
x
2
m)
ax
a1
1
x
b
2
= p
1
, bx
a
1
x
b1
2
= p
2
and
p
1
x
1
+p
2
x
2
= m. Elim.
ax
2
bx
1
=
p
1
p
2
.
Subst. in b.c.: p
1
x
1
+
b
a
p
1
x
1
= m
p
1
x
1
=
a
a+b
m, p
2
x
2
=
b
a+b
m
3
This Version of Workbook Chapter 9: September 15, 2003
9. CONSTRAINED OPTIMISATION 171
_

Worksheet 9: Constrained
Optimisation Problems
Quick Questions
(1) A consumer has utility function u(x
1
, x
2
) = 2 lnx
1
+ 3 lnx
2
, and income m = 50.
The prices of the two goods are p
1
= p
2
= 1. Use the MRS condition to determine
his consumption of the two goods. How will consumption change if the price of good
1 doubles? Comment on this result.
(2) Repeat the rst part of question 1 using the Lagrangian method and hence deter-
mine the marginal utility of income.
(3) Is the utility function u(x
1
, x
2
) = x
2
+ 3x
2
1
well-behaved? Explain your answer.
(4) A rm has production function F(K, L) = 8KL. The wage rate is 2 and the rental
rate of capital is 1. The rm wants to produce output y.
(a) What is the rms cost minimisation problem?
(b) Use the Lagrangian method to calculate its demands for labour and capital, in
terms of output, y.
(c) Evaluate the Lagrange multiplier and hence determine the rms marginal cost.
(d) What is the rms cost function C(y)?
(e) Check that you obtain the same expression for marginal cost by dierentiating
the cost function.
Longer Questions
(1) A rich student, addicted to video games, has a utility function given by U = S
1
2
N
1
2
,
where S is the number of Sega brand games he owns and N is the number of Nin-
tendo brand games he possesses (he owns machines that will allow him to play games
of either brand). Sega games cost 16 each and Nintendo games cost 36 each. The
student has disposable income of 2, 880 after he has paid his battels, and no other
interests in life.
(a) What is his utility level, assuming he is rational?
(b) Sega, realizing that their games are underpriced compared to Nintendo, raise
the price of their games to 36 as well. By how much must the students father
raise his sons allowance to maintain his utility at the original level?
(c) Comment on your answer to (b).
(2) George is a graduate student and he divides his working week between working on his
research project and teaching classes in mathematics for economists. He estimates
that his utility function for earning W by teaching classes and spending R hours
on his research is:
u(W, R) = W
3
4
R
1
4
.
He is paid 16 per hour for teaching and works for a total of 40 hours each week.
How should he divide his time between teaching and research in order to maximize
his utility?
(3) Maggie likes to consume goods and to take leisure time each day. Her utility function
is given by U =
CH
C+H
where C is the quantity of goods consumed per day and H is
172 9. CONSTRAINED OPTIMISATION
the number of hours spent at leisure each day. In order to nance her consumption
bundle Maggie works 24 H hours per day. The price of consumer goods is 1 and
the wage rate is 9 per hour.
(a) By showing that it is a CES function, or otherwise, check that the utility
function is well-behaved.
(b) Using the Lagrangian method, nd how many hours Maggie will work per day.
(c) The government decides to impose an income tax at a rate of 50% on all income.
How many hours will Maggie work now? What is her utility level? How much
tax does she pay per day?
(d) An economist advises the government that instead of setting an income tax it
would be better to charge Maggie a lump-sum tax equal to the payment she
would make if she were subject to the 50% income tax. How many hours will
Maggie work now, when the income tax is replaced by a lump-sum tax yielding
an equal amount? Compare Maggies utility under the lump-sum tax regime
with that under the income tax regime.
(4) A consumer purchases two good in quantities x
1
and x
2
; the prices of the goods are
p
1
and p
2
respectively. The consumer has a total income I available to spend on the
two goods. Suppose that the consumers preferences are represented by the utility
function
u(x
1
, x
2
) = x
1
3
1
+x
1
3
2
.
(a) Calculate the consumers demands for the two goods.
(b) Find the own-price elasticity of demand for good 1,
p
1
x
1
x
1
p
1
. Show that if p
1
= p
2
,
then this elasticity is
5
4
.
(c) Find the cross-price elasticity
p
2
x
1
x
1
p
2
when p
1
= p
2
.
(5) There are two individuals, A and B, in an economy. Each derives utility from his
consumption, C, and the fraction of his time spent on leisure, l, according to the
utility function:
U = ln(C) + ln(l)
However, A is made very unhappy if Bs consumption falls below 1 unit, and he
makes a transfer, G, to ensure that it does not. B has no concern for A. A faces a
wage rate of 10 per period, and B a wage rate of 1 per period.
(a) For what fraction of the time does each work, and how large is the transfer G?
(b) Suppose A is able to insist that B does not reduce his labour supply when he
receives the transfer. How large should it be then, and how long should A work?