Static Optimization

Mohammed Bouaddi

American University in Cairo

September 8, 2022

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Static Optimization: Unconstrained case

Outline
1. Critical values
2. The second derivative and concavity/convexity.
3. Unconstrained optimization: one-variable case.
4. Unconstrained optimization: two-variable case.
5. Second-order conditions: unconstrained optimization.
6. Constrained optimization: one-variable functions: the Lagrangian.
7. Constrained optimization: two-variable functions: the Lagrangian.
8. Second-order conditions: constrained optimization.
9. Applications.
10. About the Lagrange multiplier.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

1. Critical points
De…nition
The critical point of a function of a real variable is a point in the domain
of the function for which the …rst derivative is either equal to zero or
doen’t exixst.
Examples:
f (x ) = 2x 2 + 3, the critical point is 34 (it solves
f 0 (x ) = 4x + 3 = 0)
2x if x > 0
f (x ) = , the critical point is 0 since f 0 (x ) doen’t
2x if x 6 0
exixst at 0 (Corner point)

2 if x > 0
f 0 (x ) =
2 if x 6 0

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

De…nition
Taylor’s expansion theorem: Let n > 1 be an integer and let the function f
: R ! R be n times di¤erentiable at a point x 2 R, then there exists a
function R : R ! R such that

x x (x x )2 00
f (x ) = f (x ) + f 0 (x ) + f (x )
1! 2!
(x x ) n (n )
+... + f (x ) + R (x x )n +1
n!

where lim R (x x )n +1 = 0.
x !x

If n = 2, then

x x (x x )2 00
f (x ) = f (x ) + f 0 (x ) + f (x )
1! 2!
+R (x x )3
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
If n = 1, then

x x
f (x ) = f (x ) + f 0 (x ) + R (x x )2
1!

2. The second derivative and concavity/convexity

The curvature is an important aspect describing the shape of a function.
It can be approached using the second-order derivatives that gives the
slope of the …rst derivative at a particular value of the variable.

Univariate case: The second derivative and concavity/convexity.

De…nition
A twice di¤erentiable function f (x ) is convex if f 00 (x ) > 0 at all points on
its domain.
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
De…nition
A twice di¤erentiable function f (x ) is strictly convex if f 00 (x ) > 0 at all
points on its domain.

De…nition
A twice di¤erentiable function f (x ) is concave if f 00 (x ) 6 0 at all points
on its domain.

De…nition
A twice di¤erentiable function f (x ) is strictly concave if f 00 (x ) < 0 at all
points on its domain.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Example
Strictly convex function: f (x ) = 2x 2 3

y
40

30

20

10

-4 -2 2 4
x

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Example
Strictly concave function: f (x ) = 2x 2 + 3

x
-5 -4 -3 -2 -1 1 2 3 4 5
y
-10

-20

-30

-40

3. Unconstrained optimization: one-variable case: Maxima and

minima
Usually in mathematics we denote the optimum of a univariate function
f (x ) by x . The optimum exists at the angular point or at a point for
which the …rst derivative is zero.
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
Maxima are point at which the function attains its highest value in a
given interval or any points for which f 00 (x ) < 0.

Minima are point at which the function attains its lowest value in a given
interval or any points for which f 00 (x ) > 0.

In‡ection point: An In‡ection point is a point that neither is a maximum

nor is minimum. It is a point at which the second derivative change the
sign at x .

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Example
f (x ) = x 3 + 1

y
100

50

-5 -4 -3 -2 -1 1 2 3 4 5
x
-50

-100

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

4. Unconstrained optimization: Two-variable case: Maxima and
minima
Consider a bivariate function given by f (X ) = f (x, y ) X = (x, y )0 ,
then the gradient vector 5f and the Hessian Matrix Hf are given by:
fx f1
5f = =
fy f20

fxx fxy f11 f12

Hf = = 0
fyx fyy f21 f22

First order conditions (FOC)

A stationary point (x , y ) is a point at which the gradient is zero:
5f jX =X = 0.
The Hessian matrix, in multivariate optimization problems, plays a similar
role as the second derivative in univariate case. This comes from the
concept of de…niteness: (semi)-positive or (semi)-negative.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Examples (1)
Maximum
f (x, y ) = 2x 2 3y 2

fx 4x
5f = = =0
fy 6y

x = 0 and y = 0

The the point (x , y ) = (0, 0) is a stationary point of f (x, y ).

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

 -4 -4
-2 -2
0
-200 0

y 2-40
-60
2x
4f 4
-80
-100
-120

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Examples (2)
Minimum
f (x, y ) = 2x 2 x + 3y 2 2y + 4

fx 4x 1
5f = = =0
fy 6y 2

1 1
x = and y =
4 3

The the point (x , y ) = ( 14 , 31 ) is a stationary point of f (x, y ).

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 500
400
300
-10 z 200 -10
-5 100 -5
0
0 0

y5 5x

10 10

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Examples
saddle point

f (x, y ) = 2x 2 x 3y 2 + 2y + 4

fx 4x 1
5f = = =0
fy 6y + 2

1 1
x = and y =
4 3

The the point (x , y ) = ( 14 , 31 ) is a stationary point of f (x, y ).

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 60
-4 40 -4
20
-2 -2
0
z -200 0
-40
y 2-60 2x
4 -80 4

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

5. The Hessian matrix and concavity/convexity

fxx fxy f11 f12

Hf = = 0 0
fyx fyy f21 f22

The de…niteness of the Hessian matrix enables us to determine whether a

stationary point is a global optimum (min or max), local optimum (min or
max) or a saddle point.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Theorem

Let f (X ) be a twice di¤erentiable function with respect of each of its

arguments and suppose that X is stationary point of f (i.e.
5f jX =X = 0), then

1 If the Hessian Hf is negative de…nite, then X is a local maximum of

f . Further, if f is a strictly concave function, then X is a global
maximum of f .
2 If the Hessian Hf is positive de…nite, then X is a local minimum of
f . Further, if f is a strictly convex function, then X is a global
minimum of f .
3 If the Hessian Hf is positive inde…nite, then X is a neither a local
minimum nor a local maximum of f . The stationary point X is said
saddle point of f .

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Strict concavity/convexity of a function f :
Theorem
A twice di¤erentiable function f of many variables is:

1 Concave If the Hessian Hf is negative semide…nite at all points of its

domain.
2 Convex If the Hessian Hf is positive semide…nite at all points of its
domain.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

Examples
Find the optima of the following functions.

1 f (x, y ) = x 2 3x + 2y 2 + y 1
2 f (x, y ) = x 2 + x 2y 2 + 3y + 5
3 f (x, y ) = x 3 + x 2y 3 + 5y

Solution
1. f (x, y ) = x 2 3x + 2y 2 + y 1
The stationary points

fx 2x 3
5f = = =0
fy 4y + 1
3 1
x = and y =
2 4
The point (x , y ) = ( 32 , 1
4 ) is a stationary point of f .
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The Hessian matrix

fxx fxy 2 0
Hf = =
fyx fyy 0 4

Minors

fxx = 2>0
2 0
det = 8>0
0 4

All the minors of the Hessian matrix are positive. Thus, the point is a
local minimum and since the Hessian is the same at all points (Constant),
then this minimum is global (The function is strictly convex).

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

 2. f (x, y ) = x2 + x 2y 2 + 3y + 5
The stationary points

fx 2x + 1
5f = = =0
fy 4y + 3

1 3
x = and y =
2 4
The the point (x , y ) = ( 12 , 43 ) is a stationary point of f .
The Hessian matrix

fxx fxy 2 0
Hf = =
fyx fyy 0 4

Minors

fxx = 2<0
2 0
det = 8>0
0 4
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The minors of the Hessian matrix alternite the sign. Thus, the point is a
local maximum and since the Hessian is the same at all points (Constant),
then this maximum is global (The function is strictly concave).

3. f (x, y ) = x 3 + xy y3
The stationary points

fx0 3x 2 + y
5f = = =0
fy0 3y 2 + x
y = 3x 2

2
3 3x 2 +x = 0
3
x 27x + 1 = 0
r
3 1 1
x = 0 or x = =
27 3
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 2
1 1
y =3 02 = 0 or y = 3 =
3 3
The the points (0, 0) and ( 31 , 31 ) are stationary points of f .
The Hessian matrix
fx 3x 2 + y
5f = = =0
fy 3y 2 + x
fxx fxy 6x 1
Hf = =
fyx fyy 1 6y
Minors
fxx = 6x
6x 1
det = 36xy 1
1 6y
Note: The Hessian depends on the values of x and y .
For the …rst stationary point (0, 0), we have
fxx = 0
0 1
det = 1
1 0
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The Hessian is inde…nite. Therefore, we have a saddle point.
1 1
For the second stationary point 3, 3 , we have

fxx fxy 2 1
Hf = =
fyx fyy 1 3
Minors

fxx = 2
2 1
det = 5
1 3
The minors of the Hessian matrix alternite the sign. Thus, the point is a
local maximum.

Practice question: Find the optima (if it exists) of the following function

f (x, y ) = x3 2y 2 + xy

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

6. Constrained optimization: two-variable functions: the
Lagrangian.
We will analyse the problems similar to

min f (x, y ) or max f (x, y ) such that g (x, y ) = 0

where h (x, y ) = 0 is the constraint. Most optimization problems in

economics and business are constrained optimization ones because of the
scarsity of the ressources and may admit some contraints on the choice
variable. For example, prices and quantities are usualy positive.
Examples:
1. Utility maximization subject to a given budget
2. Production maximization subject to availability of inputs
3. Risk minimization of investment portfolio subject to vailable amount to
invest
3. Cost minimization subject to some constraints on feasibility, etc ...
4. In econometrics, we have OLS or likelihood estimators subject to some
constraints on the parameters

Example: Assume you have the following function

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 max f (x, y ) = 100x 2 50y 2 ,
and you want to …nd the optimum given that

600x + 60y 2000 = 0.

Ignoring the constraint, we can show that the point (x , y ) = (0, 0) is a

stationary point of f . In addition, the minors of the Hessian matrix
alternate the sign
fxx fxy 100 0
Hf = = .
fyx fyy 0 50

fxx = 100
100 0
det = 5000
0 50

Thus, we have a maximum. However, 600x + 60y 2000 = 2000 6= 0.

Therefore, the constraint is not satis…ed. The unconstrained optimum is
not an optimum of constrained optimization.
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
Lagrange solution
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The Lagrange solution allows us to convert the constrained optimization
problem of two variables into an unconstrained optimization problem of
three variables by adding the so called Lagrange multiplier. The
optimization problem becomes

max f (x, y )

such that

g (x, y ) = 0.

The Lagrangian function is

L (x, y , λ) = f (x, y ) λg (x, y )

The problem becomes

max L (x, y , λ)
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The …rst order conditions give

dL (x, y , λ) df (x, y ) dg (x, y )

= λ =0
dx dx dx
dL (x, y , λ) df (x, y ) dg (x, y )
= λ =0
dy dy dy
dL (x, y , λ)
= g (x, y ) = 0
dλ
Note: The last equation of the system just repeats the constrains.

Example:Cobb-Douglas production function

maxY = AK α Lβ
K ,L

given that
pK K + pL L = B Budget constraint
Lagrange function

L (K , L, λ) = AK α Lβ λ ( pK K + pL L B)
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
The …rst order conditions give

dL (K , L, λ) d AK α Lβ d ( pK K + pL L B)
= λ
dK dK dK
= αAK α 1 Lβ λpK = 0 (1)
dL (K , L, λ) d AK α Lβ d ( pK K + pL L B)
= λ
dL dL dL
= βAK α Lβ 1 λpL = 0 (2)
dL (K , L, λ)
= ( pK K + pL L B) = 0 (3)
dλ
From equations 1 and 2 we have:
1
αA (K )α (L ) β = λpK (3)
α β 1
βA (K ) (L ) = λpL (4)

Devide each side of equation 3 by the corresponding side of equation 4 we

have:
λpK αA (K )α 1 (L ) β
=
λpL βA (K )α (L ) β 1
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
Therefore we have,

pK αA (K )α 1 (L ) β αL
= 1
= (5)
pL α
βA (K ) (L ) β βK

We notice from the last equation, that the ratio of the prices is equal to
the ratio of the corresponding marginal productions at the equilibium (at
the optimum).
From equation 5, we have

β pK
L = K (6)
α pL

Substitute equation 6 in equation 3 to get

β pK
pK K + pL K =B
α pL

Therefore, we have
αB
K =
( α + β ) pK
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
and substitute in 3 to get
βB
L =
( α + β ) pL
Second-order conditions
Assumptions on global optima:
For the local optima to be global, we need the following assumptions:
1 Convexity: The objective function is convex or concave.

2 The constraint function de…nes a convex set.

The Hessian matrix of the Lagrangian function is called the bordered

Hessian. This matrix is given by
0 1
fxx λgxx fxy λgxy gx
Hf = @ fyx λ gyx fyy λgyy gy A .
gx gy 0

Theorem
If the point (x , y , λ ) is a stationary value of the constrained problem,
then:
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 1 If the determinant of the bordered Hessian is negative at the
optimum, i.e jHf j < 0, then we have a minimum.
2 If the determinant of the bordered Hessian is positive at the optimum,
i.e jHf j > 0, then we have a maximum.

Examples
Solve the following examples

Example 1: Constant elasticity of substitution function (CES function)

δ
maxY = A (πK ρ + (1 π ) Lρ ) ρ
K ,L

given that
pK K + pL L = B Budget constraint

where:
Y is the output
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
A is the total factor productivity
K and L are the intputs (capital and labor)
ρ is the substitution parameter (elasticity of substitution parameter is
e = 11ρ)
π is the share of input capital and (1 π ) is the share of input labor.
δ is the degree of homogeneity of the production function.

δ = 1 Constant return to scale

δ < 1 Decreasing return to scale
δ > 1 Increasing return to scale

If ρ approaches 1 and δ = 1, we get a linear or perfect substitutes

function.

Y = A (πK + (1 π ) L)

If ρ approaches 0 in the limit and δ = 1, we get the Cobb–Douglas

production function.
Y = AK ρ Lρ
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 If ρ approaches ∞ and δ = 1, we get the Leontief or perfect
complements production function.

Y = min (K , L)

Lagrange function
δ
L (K , L, λ) = A (πK ρ + (1 π ) Lρ ) ρ λ ( pK K + pL L B)

The …rst order conditions give

δ

dL (K , L, λ) d A (πK ρ + (1 π ) Lρ ) ρ d ( pK K + pL L B)
= λ
dK dK dK
δ
ρ 1 1
= δπK A (πK ρ + (1 π ) Lρ ) ρ λpK = 0 (1)
δ

dL (K , L, λ) d A (πK ρ + (1 π ) Lρ ) ρ
d ( pK K + pL L B)
= λ
dL dL dL
δ
ρ 1 1
= δ (1 π) L A (πK ρ + (1 π ) Lρ ) ρ λpL = 0(2)
dL (K , L, λ)
= ( pK K + pL L B) = 0 (3)
dλ
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
From equations 1 and 2 we have:
δ
1 1
δπK ρ A (πK ρ + (1 π ) Lρ ) ρ = λpK (4)
δ
1 1
δ (1 π ) Lρ A (πK ρ + (1 π ) Lρ ) ρ = λpL (5)

Devide each side of equation 3 by the corresponding side of equation 4 we

have:
δ
1
λpK δπK ρ 1 A (πK ρ + (1 π ) Lρ ) ρ
= δ
λpL 1
δ (1 π ) Lρ 1 A (πK ρ + (1 π ) Lρ ) ρ
We notice from the last equation, that the ratio of the prices is equal to
the ratio of the corresponding marginal productions at the equilibium (at
the optimum).
Therefore we have,
pK π (K ) ρ 1
= (6)
pL (1 π ) (L ) ρ 1
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
From equation 6, we have
1
πpL ρ 1
L = K (7)
( 1 π ) pK

Substitute equation 7 in equation 3 to get

1
πpL ρ 1
pK K + pL K =B
( 1 π ) pK

Therefore, we have
B
K = 1 (8)
πp L ρ 1
pK + pL (1 π )p K

and substitute in 3 to get

B B
L = 1 = 1 (9)
πp L ρ 1 (1 π )p K ρ 1
pK (1 π )p K
+ pL pK πp L + pL
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
We can rewrite 8 and 9 as

B
K = ρ
1 ρ 1
π pL
pK 1+ 1 π
ρ 1
pK

B
L = ρ
1 ρ 1
1 π pK
pL 1+ π
ρ 1
pL

Shapes of an Isoquant (Contour plot)

Leontief or perfect complements production function.

Y = min (K , L)
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 150

100
20

50
100 100
Capital

50 50 50
20

20 20

0
0 50 100 150
Labor

Cobb–Douglas production function (A = 1, ρ = 0.5)

Y = AK ρ Lρ
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 150

10
20

0
100

50
Capital

100

50

50
20

50

20 20
0
0 50 100 150
Labor

Constant elasticity of substitution function

(A = 1, π = 0.4, ρ = 0.5, δ = 1)
δ
Y = A (πK ρ + (1 π ) Lρ ) ρ
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 150

100
Capital

10
0

50
50

100
20

50 10 0
0 50
0 50 100 150
Labor

Example 2: Minimum variance portfolio

min σ2 = π 21 σ21 + π 22 σ22 + 2π 1 π 2 σ12
π 1 ,π 2 p

given that
π1 + π2 = 1 Budget constraint

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

where:
σ2p is the variance of the return of the portfolio (quantitity of risk of
the portfolio)
σ21 is the variance of the return of stock one
σ22 is the variance of the return of stock two
σ12 is the covariance between the return of stock one and the return
of stock two
π 1 and π 2 are the shares of stock one and stock two respectively
(proportions of the invested amount allocated to each stock)

Lagrange:

L = π 21 σ21 + π 22 σ22 + 2π 1 π 2 σ12 λ (π 1 + π 2 1)

F.O.C:
dL
= 2π 1 σ21 + 2π 2 σ12 λ=0 (1)
d π1
dL
= 2π 2 σ22 + 2π 1 σ12 λ=0 (2)
d π2
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 dL
= (π 1 + π 2 1) = 0 (3)
d π1
From equation 3 we have:
π2 = 1 π1 (4)
From equations 1 and 2 we have:

λ
π 1 σ21 + π 2 σ12 = (5)
2
λ
π 2 σ22 + π 1 σ12 = (6)
2
Equate 5 and 6 to get:

π 1 σ21 + π 2 σ12 = π 2 σ22 + π 1 σ12

π 1 σ21 σ12 + π 2 σ12 σ22 = 0 (7)

Substitute equation 7 in equation 1 to get:It follows that,

π 1 σ21 σ12 + (1 π 1 ) σ12 σ22 = 0

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
Solve for π 1 to get
σ22 σ12
π1 = (8)
σ21 + σ22 2σ12
Substitute 8 in 4
σ22 σ12
π2 = 1
σ21 + σ22 2σ12
σ21 σ12
=
σ21 + σ22 2σ12
Second order conditions

L = π 21 σ21 + π 22 σ22 + 2π 1 π 2 σ12 λ (π 1 + π 2 1)

dL
= 2π 1 σ21 + 2π 2 σ12 λ
d π1
dL
= 2π 2 σ22 + 2π 1 σ12 λ
d π2
dL
= ( π 1 + π 2 1)
d π1
MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1
 0 1
fxx0 0
λgxx fxy0 0
λgxy gx0
Hf = @ fyx0 λ gyx 0 fyy0 0
λgyy gy0 A
gx0 gy0 0
0 1
2σ21 λ 0 2σ12 λ 0 1
= @ 2σ12 λ 0 2σ22 λ 0 1 A
1 1 0
0 1
2σ21 2σ12 1
= @ 2σ12 2σ22 1 A
1 1 0

2σ21 2σ12 1
jHf j = 2σ12 2σ22 1 = 2σ21 2σ22 + 4σ12
1 1 0
= 2 σ21 + σ22 2σ12 < 0.

Thus, we have a minimum (jHf j < 0).

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1

About the Lagrange multiplier
Recall that the Lagrangian function is

L (x, y , λ) = f (x, y ) λg (x, y )

Therfore,
∆L = ∆f λ∆g
At the optimum
∆L = ∆f λ ∆g = 0
Thus
∆f
λ =
∆g
Hence, the lagrange multiplier gives the change of the objective function
resulting from small change of the constraint at the optimum, i.e it gives
the marginal e¤ect of the constraint on the objective function.

MB (Business School) Econ 4061 - Mathematical Economics September 8, 2022 1/1