Lecture notes based on c 2001 Michael Carter

⃝
Foundations of Mathematical Economics All rights reserved

Optimization
1 Preliminaries
1.1 Formulation
max 𝑓 (x, 𝜽)
x∈𝐺(𝜽)

where typically the feasible set is defined by

𝐺(𝜽) = {x ∈ 𝑋 : 𝑔𝑗 (x, 𝜽) ≤ 0, 𝑗 = 1, 2, . . . 𝑚}

⊳ Note: min 𝑓 = − max (−𝑓 )

⊳ Functional v. set constraints

For example, nonnegativity x ≥ 0 can be expressed as

∙ −x ≤ 0 or
∙ 𝑋 = ℜ𝑛+

1.2 Basic questions

⊳ existence (Weierstrass Theorem 2.2)

⊳ computation

⊳ characterization

⊳ sensitivity (comparative statics)

1.3 Terminology

⊳ local v. global
If 𝑓 is concave, every local optimum is a global optimum (Exercise 5.2)

⊳ strict (unique) v. nonstrict

If 𝑓 is strictly quasiconcave, every optimum is a strict global optimum
(Exercise 5.3)

⊳ interior v. boundary

⊳ necessary v. sufficient conditions

1
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

2 Unconstrained optimization
max 𝑓 (x)
x∈𝑋

2.1 Basic first-order condition (Proposition 5.1)

⊳ If x∗ is a local maximum of 𝑓 , there exists a neighborhood 𝑆 of x∗ such

that

𝑓 (x) ≈ 𝑓 (x∗ ) + ∇𝑓 (x∗ )𝑇 (x − x∗ ) ≤ 𝑓 (x∗ ) for every x ∈ 𝑆

which implies

∙ ∇𝑓 (x∗ )𝑇 (x − x∗ ) ≤ 0 for every x ∈ 𝑆

∙ ∇𝑓 (x∗ ) = 0 if x∗ is an interior point
∂𝑓 (x∗ )
– That is, 𝐷𝑥𝑖 𝑓 [x∗ ] = ∂𝑥𝑖
= 0 for every 𝑖

⊳ A maximum must be either a stationary point of 𝑓 or a boundary

point of 𝑋 or both.

𝑓 (𝑥) 𝑓 (𝑥) 𝑓 (𝑥)

𝑥 𝑥 𝑥
Stationary Boundary Both
point point

2.2 Nonnegative variables(Corollary 5.0.1)

⊳ For the nonnegativity constraint x ≥ 0, we can collapse the necessary

conditions into

∇𝑓 (x∗ ) ≤ 0 x∗ ≥ 0 ∇𝑓 (x∗ )𝑇 x∗ = 0

2
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

2.3 Basic second-order condition (interior maximum)

⊳
max 𝑓 (𝑥1 , 𝑥2 ) = 3𝑥1 𝑥2 − 𝑥31 − 𝑥32
𝑥1 ,𝑥2

Local
Saddle
maximum
point

-1
0 2
𝑥1 1
1 0
-1 𝑥2

⊳ If x∗ is a local maximum of 𝑓 , there exists a neighborhood 𝑆 of x∗ such

that

𝑓 (x) ≈ 𝑓 (x∗ ) + ∇𝑓 (x∗ )𝑇 dx + dx𝑇 𝐻𝑓 (x∗ )dx ≤ 𝑓 (x∗ )

or
∇𝑓 (x∗ )𝑇 dx + dx𝑇 𝐻𝑓 (x∗ )dx ≤ 0 for every x ∈ 𝑆
which requires that

∙ ∇𝑓 (x∗ )𝑇 dx = 0
∙ dx𝑇 𝐻𝑓 (x∗ )dx ≤ 0 for every x ∈ 𝑆.

2.4 Necessary conditions for an interior maximum

⊳ For x∗ to be an interior local maximum of 𝑓 (x) in 𝑋, it is necessary

that

1. x∗ be a stationary point of 𝑓 , that is ∇𝑓 (x∗ ) = 0 and

2. 𝑓 be locally concave at x∗ , that is 𝐻𝑓 (x∗ ) is nonpositive definite.

⊳ Examples

𝑓 (𝑥) = 𝑥2 𝑓 ′ (0) = 0 𝑓 ′′ (0) = 2

𝑓 (𝑥) = 𝑥3 𝑓 ′ (0) = 0 𝑓 ′′ (0) = 0

3
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

2.5 Sufficient conditions for an interior maximum

⊳ If

1. x∗ is a stationary point of 𝑓 , that is ∇𝑓 (x∗ ) = 0 and

2. 𝑓 is locally strictly concave at x∗ , that is 𝐻𝑓 (x∗ ) is negative defi-
nite

then x∗ is a strict local maximum of 𝑓 .

⊳ Examples

𝑓 (𝑥) = −𝑥2 𝑓 ′ (0) = 0 𝑓 ′′ (0) = −2

𝑓 (𝑥) = −𝑥4 𝑓 ′ (0) = 0 𝑓 ′′ (0) = 0

2.6 Concave maximization (Corollary 5.1.2)

⊳ Suppose 𝑓 is concave and x∗ is an interior point of 𝑋. Then x∗ is a

global maximum of 𝑓 on 𝑋 if and only if ∇𝑓 (x∗ ) = 0.

2.7 Some examples

⊳ The competitive firm (Example 5.11)

⊳ Monopoly (Example 5.12)

⊳ Least squares regression (Example 5.13)

⊳ Maximum likelihood estimation (Exercise 5.13)

4
 Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

3 Homework
1. If 𝑓 is concave and 𝐺(𝜽) is convex, every local optimum is a global
optimum (Exercise 5.2).
2. Conjecture: If 𝑓 : ℜ → ℜ has a local maximum at 𝑥∗ which is not a
strict local maximum, then 𝑓 is constant in some neighbourhood of 𝑥.
Prove or provide a counterexample.
3. Rolle’s theorem. Suppose 𝑓 ∈ 𝐶[𝑎, 𝑏] is differentiable on (𝑎, 𝑏). If
𝑓 (𝑎) = 𝑓 (𝑏), then there exists 𝑥 ∈ (𝑎, 𝑏) where 𝑓 ′ (𝑥) = 0 (Exercise 5.8).
Note: Rolle’s theorem =⇒ the mean value theorem, which provides all the
useful theorems of calculus.

Weierstrass
theorem

Mean value
theorem

L’Hôpital’s Young’s
rule theorem

Taylor’s Inverse function

theorem theorem

Implicit function
theorem

4. Solve the problem max𝑥1 ,𝑥2 𝑓 (𝑥1 , 𝑥2 ) = 3𝑥1 𝑥2 − 𝑥31 − 𝑥32 (Example 5.8).
5. Solve the problem max𝑥1 ,𝑥2 𝑓 (𝑥1 , 𝑥2 ) = 𝑥1 𝑥2 + 3𝑥2 − 𝑥21 − 𝑥22 (Exercise
5.9)
6. Extend Corollary 5.1.2 to 𝑓 pseudoconcave.

5
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

Solutions 2
1 Suppose that x∗ is a local optimum which is not a global optimum. That
is, there exists a neighborhood 𝑆 of x∗ such that

𝑓 (x∗ , 𝜽) ≥ 𝑓 (x, 𝜽) for every x ∈ 𝑆 ∩ 𝐺(𝜽)

and also another point x∗∗ ∈ 𝐺(𝜽) such that

𝑓 (x∗∗ , 𝜽) > 𝑓 (x∗ , 𝜽)

Since 𝐺(𝜽) is convex, there exists 𝛼 ∈ (0, 1) such that

𝛼x∗ + (1 − 𝛼)x∗∗ ∈ 𝑆 ∩ 𝐺(𝜽)

By concavity of 𝑓

𝑓 (𝛼x∗ + (1 − 𝛼)x∗∗ , 𝜽) ≥ 𝛼𝑓 (x∗ , 𝜽) + (1 − 𝛼)𝑓 (x∗∗ , 𝜽) > 𝑓 (x∗ , 𝜽)

contradicting the assumption that x∗ is a local optimum.

2 False. A counterexample is
{
𝑥 𝑥≤1
𝑓 (𝑥) =
1 𝑥>1

𝑓 is constant on [1, ∞), but this is not a neighbourhood of 1.

3 By the Weierstrass theorem (Theorem 2.2), 𝑓 has a maximum 𝑥∗ and a

minimum 𝑥∗ on [𝑎, 𝑏]. Either

∙ 𝑥∗ ∈ (𝑎, 𝑏) and 𝑓 ′ (𝑥∗ ) = 0 (Theorem 5.1) or

∙ 𝑥∗ ∈ (𝑎, 𝑏) and 𝑓 ′ (𝑥∗ ) = 0 (Exercise 5.7) or

∙ Both maxima and minima are boundary points, that is 𝑥∗ , 𝑥∗ ∈ {𝑎, 𝑏}

which implies that 𝑓 is constant on [𝑎, 𝑏] and therefore 𝑓 ′ (𝑥) = 0 for
every 𝑥 ∈ (𝑎, 𝑏) (Exercise 4.7).

4 The first-order conditions for a maximum are

𝐷1 𝑓 = 3𝑥2 − 3𝑥21 = 0
𝐷2 𝑓 = 3𝑥1 − 3𝑥22 = 0

1
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

or
𝑥2 = 𝑥21 𝑥1 = 𝑥22
These equations have two solutions: 𝑥1 = 𝑥2 = 0 and 𝑥1 = 𝑥2 = 1. Therefore
(0, 0) and (1, 1) are the only stationary points of 𝑓 .
The Hessian of 𝑓 is ( )
−6𝑥1 3
𝐻(x) =
3 −6𝑥2
At (1, 1) this evaluates to
( )
−6 3
𝐻(x) =
3 −6

which is negative definite. Therefore, we conclude that the function 𝑓 attains

a local maximum at (1, 1) (Corollary 5.1.1). At the other stationary point
(0, 0), the Hessian evaluates to
( )
0 3
𝐻(x) =
3 0

which is indefinite. (0, 0) is in fact a saddle point (Section 3.7.4). Hence

x∗ = (1, 1) is the unique local maximum of 𝑓 , where the function attains
the value 𝑓 (1, 1) = 1. Note however that (1, 1) is not a global maximum
of 𝑓 since for example 𝑓 (−1, −1) = 5 > 𝑓 (1, 1). In fact, 𝑓 has no global
maximum on ℜ2 since for any 𝑥2 , 𝑓 (𝑥1 , 𝑥2 ) → ∞ as 𝑥1 → −∞.

5 The first-order conditions for a maximum are

𝐷𝑥1 𝑓 (𝑥1 , 𝑥2 ) = 𝑥2 − 2𝑥1 = 0

𝐷𝑥2 𝑓 (𝑥1 , 𝑥2 ) = 𝑥1 + 3 − 2𝑥2 = 0

which have the unique solution 𝑥∗1 = 1, 𝑥∗2 = 2. (1, 2) is the only stationary
point of 𝑓 and hence the only possible candidate for a maximum. To verify
that (1, 2) satisfies the second-order condition for a maximum, we compute
the Hessian of 𝑓 ( )
−2 1
𝐻(x) =
1 −2
which is negative definite everywhere. Therefore (1, 2) is a strict local maxi-
mum of 𝑓 . Further, since 𝑓 is strictly concave (Proposition 4.1), we conclude
that (1, 2) is a strict global maximum of 𝑓 (Exercise 5.2), where it attains its
maximum value 𝑓 (1, 2) = 3 (Figure 1).

2
Lecture notes based on c 2001 Michael Carter
⃝
Foundations of Mathematical Economics All rights reserved

(1,2,3)

3
2 𝑥
2
0
1 1
𝑥1 2

Figure 1: The strictly concave function 𝑓 (𝑥1 , 𝑥2 ) = 𝑥1 𝑥2 + 3𝑥2 − 𝑥21 − 𝑥22 has
a unique global maximum.

6 Assume 𝑓 : 𝑆 → ℜ is pseudoconcave and ∇𝑓 (x∗ ) = 0. Pseudoconcavity

means that
𝑓 (x) > 𝑓 (x∗ ) =⇒ ∇𝑓 (x∗ )𝑇 (x − x∗ ) > 0
for every x, x∗ ∈ 𝑆, which is equivalent to

∇𝑓 (x∗ )𝑇 (x − x∗ ) ≤ 0 =⇒ 𝑓 (x) ≤ 𝑓 (x∗ )

Therefore stationarity

∇𝑓 (x∗ ) = 0 =⇒ ∇𝑓 (x∗ )𝑇 (x − x∗ ) ≤ 0 =⇒ 𝑓 (x) ≤ 𝑓 (x∗ ) for every x ∈ 𝑆

x∗ is a global optimum.
Conversely, if x∗ is an interior global optimum, it must be an interior local
optimum. Therefore (Theorem 5.1), ∇𝑓 (x∗ ) = 0.