Ch1part1 2019

Chapter 1: Static optimization, part 1
September 2, 2019
Motivation
Many economic analyses rely on static optimization, e.g.

production and consumption choices.
Optimization problems consist in nding some variable(s)

x = (x1 , ..., xn ) to maximize an objective function
f (x ) = f (x1 , ..., xn ) subject to some constraint on the set of
admissible (or feasible) choices S, typically representing some
bounds on available resources.
Dierent types of optimization problems
1. Unconstrained case: S is an open subset (of the

multidimensional real space) so no boundary problems occur
2. Lagrange problem: the optimization is subject to equality
constraints
3. Nonlinear programming problem: the optimization is
subject to inequality constraints
The general approach remains invariant
1. Check whether some candidates satisfy the sucient

rst-order conditions for global optimality
2. Otherwise, record stationary points using the necessary
rst-order conditions
3. and sort out local maxima, minima and saddle points using
second-order necessary and sucient conditions for local
extrema, and determine the global maximum by comparing the
corresponding values of the objective function
Unconstrained maximization
Local results (and reminder on Taylor and Hessian)
Global results (and reminder on concave functions)
Unconstrained maximization problem
Problem (A): Finding x = (x1 , x2 , ..., xn ) to maximize the

objective function f (x) = f (x1 , x2 , ..., xn ) and no restrictions are
placed on x.
We assume that f has continuous second-order derivatives.
Note: we use notations in bold for vectors of variables:

x = (x1 , x2 , ..., xn ).
Necessary rst-order conditions for local extrema
Theorem (1)
Let x∗ = (x1∗ , ..., xn∗ ) be an interior point of S at which f has partial
derivatives. A necessary condition for x∗ to be a maximum (or
a minimum) point for f is that x∗ is a stationary point for f , i.e.
fxi (x∗ ) = 0, ∀i = 1, ..., n (1)
Proof: For a small deviation dx from x∗ , the change in f is

approximated by:
df = fxi (x∗ )dxi
X
i
If fxi (x∗ ) 6= 0, then we can nd dxi such that df 6= 0.
See reminders in linear algebra appendix!
I Hessian matrix
I Taylor's expansion
I Quadratic forms
I Deniteness and minors
Necessary second-order conditions
Theorem (2)
Suppose that x∗ is a local maximum (mininum). Then the
Hessian of f evaluated at x∗ , fxx 0 (x∗ ), must be negative
(positive) semi-denite.
If the rst-order conditions are satised but the Hessian is not
semi-denite negative or positive, then x∗ is a saddle point.
Proof: Using Taylor's expansion (higher-order terms disappear when

dx is small enough), at a local maximum, we have for any dx small:
1
f (x∗ + dx) − f (x∗ ) = (dx)0 .fxx 0 (x∗ ).dx ≤ 0
2
Global results and concave functions
We prefer conditions that identify global extrema. For this we

need to place some restrictions on the shape of the function
f (x) = f (x1 , ..., xn ) that apply everywhere on the denition domain.
Using the exact Taylor's expansion at x∗ :

1
f (x) = f (x∗ ) + (x − x∗ )0 .fx (x∗ ) + (x − x∗ )0 .fxx 0 (x∗ ).(x − x∗ ) +R
2| {z }
≤0
the last term is negative (positive), and the rst-order condition

sucient for a global maximum (minimum), if we restrict to
functions with negative (positive)-semidenite Hessian everywhere.
Reminder: concavity of C 2 functions
A set S is convex if each pair of points in S can be joined by a

segment lying entirely within S, that is
∀x, y ∈ S , [x, y] ⊂ S
A function with continuous second-order derivatives (C 2 ), dened

on a convex set X , is concave (convex) if and only if its Hessian
matrix is negative (positive)-semidenite everywhere on X .
The concavity (convexity) of C 2 functions can be checked using the
principal minors of the Hessian (see result above).
Reminder: concavity of C 1 functions
The denition above applies only to C 2 functions.
A function with continuous rst-order derivatives (C 1 ), dened on

a convex set X , is concave (convex) if and only if, ∀x1 , x2 ∈ X :
f (x2 ) − f (x1 ) ≤ (≥)(x2 − x1 )0 .fx (x1 )
Geometrically, f is concave (convex) if and only if a tangent plane

to the representative surface (or graph) of f (x ) lies above (below)
this surface everywhere on the domain of denition.
Necessary conditions for optimality
Theorem (3)
Let f (x) be a concave (convex) function; then it reaches a global
maximum (minimum) at x∗ if and only if fx (x∗ ) = 0.
Proof: by concavity of f (x), ∀x,
f (x) − f (x∗ ) ≤ (x − x∗ )0 .fx (x∗ )
if fx (x∗ ) = 0, then f (x) ≤ f (x∗ ), and x ∗ is a global maximum.
Reciprocally, if x∗ is a global maximum, it is also a local one and

must satisfy the FOC.
Reminder: concavity of functions with discontinuous
derivatives (or kinks)
A function f (x) dened on a convex set X is concave (convex) if

and only if, ∀0 ≤ t ≤ 1 and ∀x1 , x2 ∈ X ,
f (t x1 + (1 − t )x2 ) ≥ (≤)tf (x1 ) + (1 − t )f (x2 )

i.e. the segment joining any two points on the graph is never above
(below) the graph.
It can easily be checked that the previous denitions imply this one.
Strict concavity and unique global extrema
A concave function may reach its global maximum (minimum) at

several points. More restrictions can be placed on the function to
guarantee the uniqueness of the global maximum (minimum). This
is done using strictly concave (convex) functions.
The same denitions as for concavity (convexity) apply using strict

inequalities and negative (positive) denite Hessian matrices.
Example 1: Find the extrema of the function:
f (x , y , z ) = x 2 + 2y 2 + 3z 2 + 2xy + 2xz
Reminder: concavity of sums of functions
1. Let f (x) be a concave (convex) function and k ≥ 0 a

constant; then kf (x) is a concave (convex) function
2. Let f (x) and g (x) be concave (convex) functions; then
f (x) + g (x) is itself a concave (convex) function
Proof:
(f + g ) (tx1 + (1 − t )x2 ) = f (tx1 + (1 − t )x2 ) + g (tx1 + (1 − t )x2 )
≥ t f ( + g )(x1 ) + (1 − t )(f + g )(x2 )
Reminder: composition of concave functions
1. Let f (x) be a concave (convex) function and F (y ) be a

concave (convex) and increasing function; then F (f (x)) is
a concave (convex) function.
2. Let f (x) be a concave (convex) function and F (y ) be a
convex (concave) and decreasing function; then F (f (x)) is
a convex (concave) function.
Proof:
F (f (tx1 + (1 − t )x2 )) ≥ F (tf (x1 ) + (1 − t )f (x2 )) ≥ tF (f (x1 )) + (1 − t )F (f (x2 ))
(F concave)
However, a limitation of concave and convex functions is that those

properties are not preserved by any increasing transformations.
2 2 2
Example 2: g (x , y , z ) = e ax +by +cz , with a, b, c nonnegative, is
convex because f (x , y , z ) = ax 2 + by 2 + cz 2 is convex as a sum of
convex functions and the function e u is increasing and convex.
Reminder: quasiconcave and quasiconvex functions
Quasiconcave functions cannot have multiple maxima (as concave

functions) and this property remains unchanged by an increasing
transformation (not true for concave functions).
Those two characteristics are very useful in economics, e.g. utility

functions (level sets, and thus preferences, are preserved by
increasing transformations).
A function f dened over a convex set S ⊂ Rn is quasiconcave if
and only if, for any number a, the upper level set
Pa = {x ∈ S : f (x) ≥ a} is convex. A function f is quasiconvex if
−f is quasiconcave, i.e. if and only if,∀a, the lower level set
Pa = {x ∈ S : f (x) ≤ a} is convex.
A concave (convex) function is also quasiconcave (quasiconvex).
Be careful: a sum of quasiconcave functions need not be

quasiconcave.
Reminder: characterization of quasiconcavity
Let f be a function of n variables dened on S ⊆ Rn . Then f is

quasiconcave if and only if either of the following equivalent
conditions is satised, ∀x, y ∈ S and ∀λ ∈ [0, 1]:
f (λx + (1 − λ)y) ≥ min(f (x), f (y))

f (x) ≥ f (y) ⇒ f (λx + (1 − λ)y) ≥ f (y)
Reminder: composition of quasiconcave functions
Let S ⊂ Rn be a convex set, f : S → R and F a function of one

variable whose domain includes f (S ). Then:
1. Let f be a quasiconcave (quasiconvex) and F an increasing
function; then F (f (x)) is quasiconcave (quasiconvex);
2. Let f be a quasiconcave (quasiconvex) and F a decreasing
function; then F (f (x)) is quasiconvex (quasiconcave).
Proof: Assume F (f ) is not quasiconcave. Then there is u and v

such that:
F (f (λu + (1 − λ)v ) < F (f (u )) and F (f (λu + (1 − λ)v ) < F (f (v ))
f (λu + (1 − λ)v ) < f (u ) and f (λu + (1 − λ)v ) < f (v )
which is contradictory with f quasiconcave.

Reminder: homogeneous functions
For specic functions, concavity and quasiconcavity are equivalent...
A function f dened on a cone K is homogeneous of degree q if

f (t x) = t q f (x) for all x ∈ K and all t > 0. (A set K ∈ Rn is a
cone if t x ∈ K whenever x ∈ K and t > 0.)
Let f be a function dened on a convex cone K ⊂ Rn . Suppose f
is quasiconcave and homogeneous of degree q ∈ ]0, 1], that
f (0) = 0 and that f (x) > 0 for all x 6= 0 in K . Then f is concave.
Cobb-Douglas functions
Theorem (4)
The Cobb-Douglas function f (x ) = Ax1a1 ...xnan dened for all
x1 > 0, ..., xn > 0, with A, a1 , ..., an positive constants is
homogeneous of degree a = a1 + ... + an , and:
1. quasiconcave for all a1 , ..., an
2. concave for all a ≤ 1
3. strictly concave for all a < 1
Proof: Consider lnf (x ) = lnA + a1 lnx1 + ... + an lnxn .
As a sum of concave functions, lnf is concave and quasiconcave.
As f (x ) = e lnf (x ) , with u → e u increasing, f is quasiconcave.
f is homogeneous of degree a and f (0) = 0. If a ≤ 1, using the
theorem above, f is concave. For a < 1, f is strictly concave
(shown by checking the leading principal minors of the Hessian).
For a > 1, f is not concave (along the ray x1 = ... = xn = x, f is
strictly convex).
Generalized CES function
Theorem (5)
The constant elasticity of substitution (CES) function
µ
f (x ) = A(δ1 x1−ρ + ... + δn xn−ρ )− ρ dened for x1 > 0, ..., xn > 0,
with A > 0, µ > 0, ρ 6= 0, δi > 0, i = 1...n, is homogeneous of
degree µ, and
1. quasiconvex for ρ ≤ −1, quasiconcave for ρ ≥ −1
2. concave for 0 < µ ≤ 1, ρ ≥ −1
3. strictly concave for 0 < µ < 1, ρ > −1
Proof: f (x ) = Au −µ/ρ , with u = δ1 x1−ρ + ... + δn xn−ρ .
If ρ ≤ −1, u is convex and quasiconvex as a positive sum of convex
functions. As u → Au −µ/ρ is increasing, f is quasiconvex.
If ρ ∈] − 1, 0], u is concave and the transformation increasing, so f
is quasiconcave.
If ρ > 0, u is convex and the transformation decreasing, so f is
quasiconcave. f is homogeneous of degree µ. By the theorem
above, if 0 < µ ≤ 1 and ρ > −1, f is concave.
The last point is shown by an extension of the theorem to strict
concavity.
Summary of main concepts
1. Necessary rst-order conditions for local extrema: zero rst

derivatives
2. Necessary second-order conditions for local extrema and saddle
points: deniteness of the Hessian
3. Sucient conditions for local extrema: zero rst derivatives and
deniteness of the Hessian
4. Necessary conditions for global extrema: concavity/convexity of the
objective (3 characterizations) and zero rst derivatives
5. Properties of concave functions: not preserved by any increasing
transformation
6. Quasiconcave functions (which include concave functions) have this
property, examples of Cobb-Douglas and CES functions
7. Quasiconcavity is equivalent to concavity for a subset of functions
(dened on a cone, positive, homog. of degree ≤ 1).

Ch1part1 2019

Uploaded by

Copyright:

Available Formats

Ch1part1 2019

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ch1part1 2019

Uploaded by

Copyright:

Available Formats

Chapter 1: Static optimization, part 1

Many economic analyses rely on static optimization, e.g.

Optimization problems consist in nding some variable(s)

1. Unconstrained case: S is an open subset (of the

1. Check whether some candidates satisfy the sucient

Problem (A): Finding x = (x1 , x2 , ..., xn ) to maximize the

We assume that f has continuous second-order derivatives.

Note: we use notations in bold for vectors of variables:

fxi (x∗ ) = 0, ∀i = 1, ..., n (1)

Proof: For a small deviation dx from x∗ , the change in f is

Proof: Using Taylor's expansion (higher-order terms disappear when

We prefer conditions that identify global extrema. For this we

Using the exact Taylor's expansion at x∗ :

the last term is negative (positive), and the rst-order condition

A set S is convex if each pair of points in S can be joined by a

A function with continuous second-order derivatives (C 2 ), dened

The denition above applies only to C 2 functions.

A function with continuous rst-order derivatives (C 1 ), dened on

f (x2 ) − f (x1 ) ≤ (≥)(x2 − x1 )0 .fx (x1 )

Geometrically, f is concave (convex) if and only if a tangent plane

Proof: by concavity of f (x), ∀x,

f (x) − f (x∗ ) ≤ (x − x∗ )0 .fx (x∗ )

if fx (x∗ ) = 0, then f (x) ≤ f (x∗ ), and x ∗ is a global maximum.

Reciprocally, if x∗ is a global maximum, it is also a local one and

A function f (x) dened on a convex set X is concave (convex) if

f (t x1 + (1 − t )x2 ) ≥ (≤)tf (x1 ) + (1 − t )f (x2 )

A concave function may reach its global maximum (minimum) at

The same denitions as for concavity (convexity) apply using strict

1. Let f (x) be a concave (convex) function and k ≥ 0 a

1. Let f (x) be a concave (convex) function and F (y ) be a

However, a limitation of concave and convex functions is that those

Quasiconcave functions cannot have multiple maxima (as concave

Those two characteristics are very useful in economics, e.g. utility

A concave (convex) function is also quasiconcave (quasiconvex).

Be careful: a sum of quasiconcave functions need not be

Let f be a function of n variables dened on S ⊆ Rn . Then f is

f (λx + (1 − λ)y) ≥ min(f (x), f (y))

Let S ⊂ Rn be a convex set, f : S → R and F a function of one

Proof: Assume F (f ) is not quasiconcave. Then there is u and v

which is contradictory with f quasiconcave.

For specic functions, concavity and quasiconcavity are equivalent...

A function f dened on a cone K is homogeneous of degree q if

1. Necessary rst-order conditions for local extrema: zero rst

You might also like

Optimization problems consist in nding some variable(s)

1. Check whether some candidates satisfy the sucient

the last term is negative (positive), and the rst-order condition

A function with continuous second-order derivatives (C 2 ), dened

The denition above applies only to C 2 functions.

A function with continuous rst-order derivatives (C 1 ), dened on

A function f (x) dened on a convex set X is concave (convex) if

The same denitions as for concavity (convexity) apply using strict

Let f be a function of n variables dened on S ⊆ Rn . Then f is

For specic functions, concavity and quasiconcavity are equivalent...

A function f dened on a cone K is homogeneous of degree q if

1. Necessary rst-order conditions for local extrema: zero rst