Lecture notes based on

Foundations of Mathematical Economics

c 2001 Michael Carter

All rights reserved

Constraint qualication and sucient

conditions
1 Constraint qualication
We derived the Kuhn-Tucker theorem by assuming regularity of the binding constraints. With the possibility of corner solutions, regularity is too
stringent. The regularity condition will not necessarily apply throughout
the constraint set. Verifying the regularity condition requires knowing the
optimal solution, but the rst-order conditions will not identify the optimal
solution unless the regularity condition is satised Catch 22.
To illustrate, consider the problem
max 1
1 ,2

subject to 2 (1 1 )3 0
2 0
2

1()
1

2()
The solution is obviously (1, 0), but this does not satisfy the Kuhn-Tucker
conditions. The gradients of the two constraints at (1, 0) are (0, 1) and (0, 1)
which are linearly dependent. The constraints are not regular at the optimum
(1, 0). The problem is that the set
(x ) = {dx : (x ) dx 0,

= 1, 2} = {(1 , 0)}

does not represent the set of feasible perturbations (x ) = {(1 , 0) : 1

1}. If we add an additional constraint 3 (1 , 2 ) = 1 + 2 1, the three
gradients are still linearly dependent, but now
(x ) = {dx : (x ) dx 0,

= 1, 2, 3} = (x )

and the optimum (1, 0) satises the Kuhn-Tucker conditions. We seek conditions on the constraints g such that the set (x ) correctly represents the set
of feasible perturbations (x ), which is known as the cone of tangents.
1

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Foundations of Mathematical Economics

1.1 Constraint qualication conditions (Theorem 5.4)

Suppose that x is a local solution of
max (x) subject to (x) 0
x

at which the binding constraints (x ) satisfy one of the following constraint

qualication conditions.
Concave CQ is concave for every (x )
such that
Pseudoconvex CQ is pseudoconvex and there exists x

x) < 0 for every (x )

(

Quasiconvex CQ is quasiconvex, (x ) = 0 and there exists x

such that (
x) < 0 for every (x )
Regularity The set { (x ) : (x ) } is linearly independent
Then the Kuhn-Tucker conditions are necessary for an optimal solution.
The most common constraint qualication conditions encountered are
linear = concave (e.g. linear programming)
such that (
Slater condition: convex and there exists x
x) <
0 for every
1.2 Constraint qualication with nonnegative variables (Corollary 5.4.1)
Provided the binding constraints (x ) in the problem
max (x) subject to g(x) 0
x0

satisfy any one of the following constraint qualication conditions:

Concave CQ
Pseudoconvex CQ
Quasiconvex CQ
the Kuhn-Tucker conditions are necessary for an optimal solution.
Proof. The problem can be specied as
max (x)
x

subject to g(x) 0
h(x) = x 0
2

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Foundations of Mathematical Economics

We note that h is linear, and is therefore both concave and convex. Further
h[x] = 0 for every x. Therefore, if g satises one of the three constraint
qualication conditions, so does the combined constraint (g, h). By Theorem
5.4, the Kuhn-Tucker conditions are necessary for a local optimum.

Example 5.41: The consumers problem The consumers problem

max (x)
x0

subject to p x
has one functional constraint (x) = p x and inequality constraints
(x) = 0, the gradients of which are
= p

= e ,

= 1, 2, . . . ,

where e is the unit vector (Example 1.79). Provided all prices are positive
p > 0, it is clear that these are linearly independent and the regularity
condition of Corollary 5.3.2 is always satised. However, it is easier to appeal
directly to Corollary 5.4.1, and observe that the budget constraint (x) =
p x is linear and therefore concave.

2 Sucient conditions
We know that the KT conditions are sucient when is concave and is
convex. The following is a signcant generalization (e.g. consumer theory).
2.1 Sucient conditions for a global optimum (Theorem 5.5)
Suppose that x satises the KT conditions and
is pseudoconcave
is quasiconvex
Then x is a global maximum.
Proof. For every
either = 0 which implies (x ) (x x ) = 0 for every x
or (x ) = 0 and therefore (x) 0 = (x ) for every x = {x :
(x) 0, = 1, 2, . . . , }. Quasiconvexity implies
(x ) (x x ) 0 for every x
and since 0
(x ) (x x ) 0 for every x
3

Lecture notes based on

Foundations of Mathematical Economics

c 2001 Michael Carter

All rights reserved

The KT condition is

(x ) =

(x )

=1

and therefore

(x ) (x x ) =

(x ) (x x ) 0

=1

Since is pseudoconcave, this implies that (x ) (x)

2.2 Arrow-Enthoven theorem (Corollary 5.5.1)

Suppose that x satises the KT conditions and
is quasiconcave and (x ) = 0
is quasiconvex
Then x is a global maximum.
2.3 Necessary and sucient conditions (Corollary 5.5.3)
Suppose that
is quasiconcave and quasiconvex
(x ) = 0 and (x ) = 0 for every (x )
such that (
x) < 0 for every (x )
there exists x
Then the KT conditions are necessary and sucient for a global optimum.
Example 5.43: The consumers problem Convex preferences and nonsatiation imply quasiconcave and (x) = 0 for every x . Therefore
the KT conditions
(x ) = p
are necessary and sucient for utility maximization.

Lecture notes based on

Foundations of Mathematical Economics

c 2001 Michael Carter

All rights reserved

Example: Linear programming (Section 5.4.4) A linear programming problem is a special case of the general constrained optimization problem
max (x)
x0

(1)

subject to (x) 0
in which both the objective function and the constraint function g are
linear. Consequently, the Kuhn-Tucker conditions are both necessary and
sucient for a global optimum. The simplex algorithm is an ecient algorithm for solving the Kuhn-Tucker conditions.

3 Homework
1. Suppose that x is a local solution of
max (x) subject to (x) 0
x

at which the binding constraints (x ) satisfy one of the above constraint qualication conditions, so that x satises the Kuhn-Tucker
conditions

(x ) =

(x ) and (x ) = 0 = 1, 2 . . . ,

=1

Show that the Lagrange multipliers 1 , 2 , . . . , are unique if and

only if (x ) satises the Regularity condition, that is { (x ) :
(x ) } is linearly independent.
2. Suppose that a rm has contracted with its union to hire at least
units of labour at rate 1 per unit. It can also hire non-union labour
at 2 < 1 per hour. Assume that labour is the only input. Union and
non-union labour are equally productive, with diminishing marginal
product. Output is sold at a xed price .
(a) Derive and interpret the rst-order conditions for maximizing prot.
(Note: Although the optimal solution is obvious, please show how
it can be derived from the rst-order conditions.)
(b) Are these conditions necessary for a solution?
(c) Are they sucient to identify a global optimum?

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Foundations of Mathematical Economics

Solutions 5
1 Since = 0 for every
/ (x ), the Kuhn-Tucker conditions imply

(x ) =
(x )
(x )

If (x ), (x ) are independent, then the are unique (Exercise

for
1.137). Conversely, if there exist 1 , 2 , . . . , such that with =
some and

(x )
(x ) =
(x )

then

(x ) (x ) =

( ) (x ) = 0

(x )

which implies that (x ), (x ) are dependent (Exercise 1.133).

2

(a) Let 1 denote the quantity of union labour and 2 the quantity of
non-union labour hired by the rm. The rms optimisation problem
is
max (1 + 2 ) 1 1 2 2
1 ,2 0

which can be written as

max

1 and 2 0

(1 + 2 ) 1 1 2 2

subject to (1 ) = 1 0
Forming the Lagrangean
(1 , 2 , ) = (1 + 2 ) 1 1 2 2 ( 1 )
the Kuhn-Tucker conditions for an optimum are
1 = (1 + 2 ) 1 + = 0

(1)

2 = (1 + 2 ) 2 0 2 0 () 2 2 = 0 (2)
1

( 1 ) = 0

Lecture notes based on

Foundations of Mathematical Economics

c 2001 Michael Carter

All rights reserved

There are two cases:

2 > 0 (1) and (2) imply
1 + = 2 = > 0 = 1 =
and

(2 + ) = 2

2 = 0 (2) implies
(1 ) 2 < 1 = > 0 = 1 =
In both cases, the rm hires exactly units of union labour (1 ). If
() > 2 , then the rm hires additional units of non-union labour
(2 ).
(b) Since the constraint function (1 ) = 1 is linear, it is satises the
CQ condition. The Kuhn-Tucker conditions are necessary.
(c) Diminishing marginal product means that the production function is
concave. Therefore the objective function (1 + 2 ) 1 1 2 2 is
concave. Since the constraint function is convex (linear), the KuhnTucker conditions are also sucient.