2.
3 Optimization with Inequality Constraints: Kuhn-Tucker Conditions
Optimization with inequality constraints or nonnegativity constraints
Example:
) , , ( max
3 2 1
x x x f z =
s.t. b x x x g ) , , (
3 2 1
0 , ,
3 2 1
x x x
Kuhn-Tucker Conditions
(1) Nonnegativeiy Restriction
Optimization problem only with nonnegativity constraints
(A) one choice variable
) ( max
1
x f z =
s.t. 0
1
x
Three possible critical points
(i)
*
1
x
1
x
0
*
1
> x
0
) (
1
1
=
dx
x df
z
A
Point A
: a local maximum occurs in
the interior of the feasible
set
: interior solution ( )
F.O.C:
(ii)
z
B
Point B
: a local maximum occurs
on the vertical axis
: boundary solution ( ) 0
*
1
= x
F.O.C: 0
) (
1
1
=
dx
x df
1
x
*
1
x
(iii)
1
x
0
*
1
= x
0
) (
1
1
<
dx
x df
z
Point C
: a local maximum occurs
on the vertical axis
C
: boundary solution ( )
F.O.C:
*
1
x =0
Three conditions:
(i) and (Point A) 0 ) ( '
1
= x f 0
1
> x
or (ii) and 0 ) ( '
1
= x f 0
1
= x (Point B)
or (iii) and 0 ) ( '
1
< x f 0
1
= x (Point C)
Combine three conditions to form a single statement
: and 0 ) ( '
1
x f 0 ) ( '
1 1
= x f x and 0
1
x
(B) n choice variables
) , , , ( max
2 1 n
x x x f z L =
s.t. 0 , , ,
2 1
n
x x x L
Khun-Tucker Conditions
(i) and (ii) 0 ) ( '
i
x f 0 ) ( ' =
i i
x f x and (iii) for i = 1, 2, , n 0
i
x
(2) Inequality Constraints
Derive Kuhn-Tucker Conditions for optimization with inequality constraints intuitively
using Khun-Tucker Conditions for optimization with nonnegativity constraints. To do that,
introduce additional variables, (read Chiang pp724-730).
i
s
n-variable m-constraint case
max ) , , , (
2 1 n
x x x f z L =
s.t.
1 2 1
1
) , , , ( r x x x g
n
L
M
m n
m
r x x x g ) , , , (
2 1
L
0 , , ,
2 1
n
x x x L
( ) | |
n
i
i
m
i
i n
x x x g r x x x f L , , , ) , , , (
2 1
1
2 1
L L + =
=
Kuhn-Tucker Conditions
=
=
m
i
i
j i j
j
g f
x
L
1
0 , , 0
j
x 0 =
j
j
x
L
x
0 =
i
i
i
g r
L
, 0
i
, 0 =
i
i
L
for i and , m , , 2 , 1 L = n j , , 2 , 1 L =
where
j
j
x
f
f
= and
j
i
i
j
x
g
g
=
Complementary Slackness Conditions
= <
>
0 0
0 0
j
j
j
j
x
x
L
x
L
x
0 =
j
j
x
L
x Q , 0
j
x
L
, 0
j
x
= >
>
0 0
0 0
i
i
i
i
L
L
0 =
i
i
L
Q , 0
i
L
, 0
i
,
(3) A Minimization Problem
min ) , , , (
2 1 n
x x x f c L =
s.t.
1 2 1
1
) , , , ( r x x x g
n
L
M
m n
m
r x x x g ) , , , (
2 1
L
0 , , ,
2 1
n
x x x L
max ) , , , (
2 1 n
x x x f c L =
s.t.
1 2 1
1
) , , , ( r x x x g
n
L
M
m n
m
r x x x g ) , , , (
2 1
L
0 , , ,
2 1
n
x x x L
(4) Kuhn-Tucker Sufficiency Theorem
Given the problem
max ) , , , (
2 1 n
x x x f z L =
s.t.
1 2 1
1
) , , , ( r x x x g
n
L
M
m n
m
r x x x g ) , , , (
2 1
L
0 , , ,
2 1
n
x x x L
if the following conditions are satisfied:
(i) the objective function is differentiable and concave in the
nonnegative orthant
) , , , (
2 1 n
x x x f L
(ii) each constraint function is differentiable and convex in the
nonnegative orthant
) , , , (
2 1 n
i
x x x g L
(iii) the point satisfies the Kuhn-Tucker maximum conditions
* *
2
*
1
, , ,
n
x x x L
then gives a global maximum of
* *
2
*
1
, , ,
n
x x x L ) , , , (
2 1 n
x x x f z L = .
Concavity and Convexity
A function ) (x f is if and only if for any pair of distinct points u and v in
the domain of
convex
concave
) (x f , and for 0 1 < < , . | | v u f v f u f ) 1 ( ) ( ) 1 ( ) ( +
+
Notes
1. If ) (x f is a linear function, then it is a concave function as well as a convex
function.
2. If ) (x f is a concave function, then ) (x f is a convex function and vice versa.
Theorem
If ) (x f and ) (x g are both concave (convex) functions, then ) ( ) ( x g x f + is also a
concave (convex) function.
Theorem
A twice continuously differentiable function ) , , , (
2 1 n
x x x f z L = is ,
convex
concave
if and only if , ie
0
0
2
2
z d
z d
( )
0 , , 0 , 0 0
0 1 , , 0 , 0 , 0
3 2 1
3 2 1
n
n
n
H H H H
H H H H
L
L
where and
(
(
(
(
(
(
=
nn n n
n
n
f f f
f f f
f f f
H
, , ,
, , ,
, , ,
12 1
2 22 21
1 12 11
L
M M M
L
L
i j
ij
x
f
x
f
= .
Ex) : concave or covex?
2
2
2
1 2 1
) 2 ( ) 3 ( ) , ( = x x x x f
Answer: Check H. Concave function
Ex) (Dowling, 13.36)
max 13 4 96 2 64 ) , (
2 2
+ = y y x x y x f
s.t. 36 + y x
0 , 0 y x
derive Kuhn-Tucker Conditions
0 * , 12 * , 16 * = = = y x
Check whether Kuhn-Tucker sufficiency theorem is applicable.
(5) Eonomic Application: Sales Maximizing Firm
The goal of a firm is maximizing its sales (total revenue) and keeping that the profit level
does not fall below a certain minimum profit which is below maximum profit.
The problem of a firm
max ( ) q R
s.t.
0
) ( ) ( = q C q R
0
) ( ) ( + = q C q R
0 q
where is differentiable and concave and is differentiable and convex (i.e. the
constraint function is differentiable and convex).
) (q R ) (q C
) ( ) ( q C q R +
R, C
C(q)
If minimum profit =
0
, output = Q
0
and revenue = R
0
R(q)
q
0
R
R
0
R*
q
Q* Q
R
Q
0
Feasible Region
For the problem
| | ) (q R L + = ) ( ) (
0
q C q R +
uhn-Tucker Conditions K
0 ) ( ' ) ( '
q C q
q
) ( ' + =
R q R
L
(1)
(2) 0 q
0 =
q
L
q (3)
0 ) ( ) (
0
+ =
q C q R
L
(4)
0
=
(5)
0
L
(6)
Try
0) =0 C(0) >0
i) 0 = q
R( ,
=
C
L
0 ) (
0
<
violate (4)
ii) Try 0 > q
L
0 ) ( ' ) ( ' ) ( ' = + =
q C q R q R
q
) ( '
1
) ( ' q C q R
+
=
(a) 0 =
) 0 ( ' = q R
not feasible under a
maximize total revenue
ssumptions
(b) 0 >
= 0 ) ( ) (
0
= +
q C q R
L
by complementary slackness condition
optimal solution: a firm attempts to earn
0
, the minimum profit required.
