Economics of Imperfect Markets

(Lecture notes)

Prof. Hervé Crès

NYUAD - Spring 2023

Contents

I Economics of Perfect Markets 1

1 Perfect markets: Positive analysis 2

1.1 The benchmark economy . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Consumers and preferences . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Utility maximization . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.3 Offer curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Pure exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 The Edgeworth box . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.2 Walrasian equilibrium . . . . . . . . . . . . . . . . . . . . . . 12
1.2.3 Intertemporal choice . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Producers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Production sets . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Profit maximization . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 The general model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Endowments and allocations . . . . . . . . . . . . . . . . . . . 18
1.4.2 Returns on portfolios . . . . . . . . . . . . . . . . . . . . . . . 18
1.4.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.4 Shareholders are unanimous . . . . . . . . . . . . . . . . . . . 20
1.4.5 Value vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Quadratic utilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

i
CONTENTS ii

II Imperfect Financial Markets 27

III Externalities and Public Goods 28

IV Market Power 29

V Information 30

VI Political Economy of Imperfect Markets 31

A Mathematical appendix 32
A.1 Sets and relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
A.2 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
A.3 Linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
A.4 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
A.5 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
A.6 Convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
 Part I

Economics of Perfect Markets

1
Chapter 1

Perfect markets: Positive analysis

In this chapter, we study the essence of the market mechanism: the exchange of
goods or services between free agents. Agents can be individual (e.g., consumers)
or collective (e.g., producers, the state).
The agents will be modeled in isolation of the others in Section 1.1 and Section
1.3, for consumers and producers respectively. We model the interaction between
agents in Section 1.2 for the case of a pure exchange economy, and in Section 1.4
in the general case with concumption and production. Section 1.5 illustrates the
material for the important class of quadratic preferences.
Section 1.1 introduces the fundamentals of economic choice by individual agents
(consumers) in isolation of others. It answers to two questions:

ˆ How do individual agents assign value to goods and services? The notion of
preferences, and their axiomatic foundations, and the notion of utility that
stems from preferences are presented in subsection 1.1.1.

ˆ What do individual agents decide when facing a constrained environment?

The notion of demand is presented in subsection 1.1.2 and that of an offer
curve in subsection 1.1.3.

Section 1.2 introduces the fundamentals of economic exchange between indi-

vidual agents (consumers). It proceeds in two steps:

ˆ For two individual agents, in the Edgeworth box, the geometry of net de-
mands of both sides are represented and compared in subsection 1.2.1.

ˆ The notion of equilibrium is introduced and studied in subsection 1.2.2.

2
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 3

Section 1.3 introduces the fundamentals of economic choice by producers. It

proceeds in two steps:

ˆ It models the physical and technological constraints faced by a producer through

the notion of production set in subsection 1.3.1.

ˆ Then it introduces one possible rationale for production choice in subsection

1.3.2: that of profit maximization.

Section 1.4 introduces the general model of exchange between consumers and
producers on a perfect market.

ˆ Endowments and property rights on the (collectively owned) production is

presented in subsection 1.4.1.

ˆ Then the rationale underpinning the objective of profit maximization is stur-

died in subsection 1.4.2.

ˆ The notion of equilibrium of the pure exchange economy is extended to this

production economy in subsection 1.4.3.

ˆ A notion of general equilibrium is introduced in subsection 1.4.4 based on

an analysis of collective decision.

ˆ Finally, subsection 1.4.5 introduces the notion of value vector characterizing

what values agent give to bundles of goods and services, a notion that will re-
veal crucial for the normative analysis of Chapter 2, and for the understanding
of the political economy of the firm.

1.1 The benchmark economy

The economy is represented in a stylized manner by a traditional general equilibrium
model, in line with Arrow & Debreu (1954).

1.1.1 Consumers and preferences

There are L goods in the economy, indexed from 1 to L, where L = {1, . . . , L}. We
assume that quantities of goods are measurable and divisible. A bundle of goods is
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 4

represented by a vector x = (x1 , x2 , . . . , xL ) ∈ RL , where xℓ is the quantity of good

ℓ in the bundle x. (Think about a negative quantity as a debt.)
There are I consumers, indexed from 1 to I, where I = {1, . . . , I}. We claim
that consumers is able to relate bundles to one another, which is easy to represent
in logical terms by the notion of binary relation over the space/set of bundles.
A consumer is characterized by some preference over the space of bundles. A
preference is a binary relation over RL , denoted ⪰, that satisfies the following prop-
erties:

ˆ completeness: ∀x, x′ ∈ RL , x ⪰ x′ or x′ ⪰ x or both;

ˆ transitivity: ∀x, x′ , x′′ ∈ RL , x ⪰ x′ and x′ ⪰ x′′ ⇒ x ⪰ x′′ .

Such a preference defines a strict preference, denoted ≻:

x ≻ x′ ⇐⇒ x ⪰ x′ but x′ ⪰̸ x ,

and an indifference relation (or equivalence relation - see Definition A.2), denoted
∼:
x ∼ x′ ⇐⇒ x ⪰ x′ and x′ ⪰ x.

For a bundle x ∈ RL , the upper contour set P (x) = {x′ ∈ RL | x′ ⪰ x} is

the set of bundles that are preferred to x by preference ⪰; the lower contour set
Q(x) = {x′ ∈ RL | x ⪰ x′ } is the set of bundles to which x is preferred; the strict
upper contour set P̊ (x) = {x′ ∈ RL | x′ ≻ x} is the set of bundles strictly preferred
to x; and I(x) = {x′ ∈ RL | x′ ∼ x} is the set of bundles equivalent to x (the
indifference classes). We have

P (x) = P̊ (x) ∪ I(x) and I(x) = P (x) ∩ Q(x)

Claim 1.1 The indifference classes form a partition of RL .

Proof: See Definition A.1 and Theorem A.1. 2

Based on this claim, the assumptions of completeness and transitivity of the pref-
erence are the essence of rationality. In some cases, they are enough to ensure
existence of a utility function that represents the consumer’s behavior.

Exercise 1.1 Let ⪰ be a preference (complete and transitive binary relation) over
a finite set X ⊆ RL . Prove that there is a utility function u : X → R that represents
it (i.e., such that x ⪰ x′ ⇐⇒ u(x) ≥ u(x′ )).
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 5

In order to have utility functions on the whole consumption set RL , we will also
assume that the preference ⪰ satisfies:

ˆ continuity: ∀x ∈ RL , P (x) and Q(x) are closed sets;

ˆ strict monotonicity: x ≥ x′ 1
and x ̸= x′ ⇒ x ≻ x′ .

Under these assumptions, there exists a utility function over RL that represents ⪰.

Theorem 1.1 [Representation theorem] Consider a preference relation ⪰ that is

complete, transitive, continuous and strictly monotonic. There exists a continuous
function u : RL → R such that

∀x, x′ ∈ RL , x ⪰ x′ ⇐⇒ u(x) ≥ u(x′ ) .

Proof: See Appendix. 2

Note that u is also strictly increasing:

x > x′ =⇒ u(x) > u(x′ ).

In most economic analysis, one further assumptions is made about consumers’

preferences:

ˆ (strict) convexity: ∀x ∈ RL , P (x) is strictly convex.

The strict convexity of preferences translates into concavity of the utility function
that represents it.

Claim 1.2 The preference ⪰ is strictly convex if and only if the utility function u
strictly quasi concave.

Proof: See Definition A.22 and Proposition A.13. 2

Finally, and mainly for the sake of mathematical convenience, we will assume:

ˆ differentiability: the utility function u is differentiable;

1
Here ≥ is the natural (partial) order on RL ; x ≥ x′ means that x is larger than x′ coordinate
by coordinate: ∀ℓ ∈ L, xℓ ≥ x′ℓ . If x ≥ x′ and x ̸= x′ , we write x ≫ x′ .
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 6

ˆ boundedness from below: ∀x ∈ RL the set P (x) is bounded from below:

∃b(x) ∈ RL , ∀x′ ∈ P (x), b(x) ≤ x′ .

From now on, we will represent consumer behavior by utility maximization. But
the above argumentation stresses that this comes as a theorem; what is assumed is
just that agent can relate bundles to one another in ways which satisfy some prop-
erties that characterize economic rationality. The latter properties are thoroughly
discussed in economic science; this is the main object of Behavioral Economics.

1.1.2 Utility maximization

A consumer, characterized by her utility function and endowed with some wealth,
w, maximizes her utility with respect to perceived constraints.
We only consider here the budget constraint. It is represented by the inequality:
L
X
p·x= p ℓ xℓ ≤ w ,
ℓ=1

where p = (p1 , p2 , . . . , pL ) ∈ RL++ is the price vector. This constraint defines a set,
from which the consumer can choose.
The consumer maximizes her objective function, u, subject to the constraint.
She therefore solves the problem:

Max u(x)
x∈R L (1.1)
s. t. p·x≤w

Given strict monotonicity, if there is a solution to the consumer’s problem, it

must ly on the budget line, and not strictly inside the budget set.

Claim 1.3 At the maximum the constraint is binding:

p·x=w .

Proof: Suppose to the contrary that the maximum x∗ is such that p · x∗ < w. There
exists δ ≫ 0 such that p · x∗ + p · δ = p · (x∗ + δ) < w. Hence x∗ + δ is affordable;
but it brings a strictly higher utility level: u(x∗ + δ) > u(x∗ ) − a contradiction to
x∗ being the maximum. 2
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 7

Given continuity, there exists a solution to the consumer’s problem. Given con-
vexity, this solution is unique. This allows to define the an individual demand
function.

Proposition 1.1 The optimization program of the consumer has a unique solution

x∗ = f (p, w) .

The mapping f : RL++ × R −→ RL is the individual demand function.

Proof: Take x̄ on the budget constraint: p· x̄ = w. We know that if there is a solution

to the optimization problem, it is in P (x̄). Hence we do not alter the optimization
problem by adding x ∈ P (x̄) to the constraint:

Max u(x)
x∈R L
(
p·x=w
s. t.
x ∈ P (x̄)

Denote B(p, w, x̄) this new budget set and b̄ = b(x̄). Given boundedness from
below, any x ∈ B(p, w, x̄) is such that b(x̄) ≤ x, hence B(p, w, x̄) is bounded from
P
below. Moreover for all ℓ ∈ L, pℓ b̄ℓ ≤ pℓ xℓ hence p1 x1 + ℓ̸=1 pℓ b̄ℓ ≤ p · x ≤
P
w, therefore x1 ≤ (w − ℓ̸=1 pℓ b̄ℓ )/p1 ; and by the same argument xk ≤ (w −
P
ℓ̸=k pℓ b̄ℓ )/pk for all k ∈ L: B(p, w, x̄) is bounded from above. Hence bounded.
We know in addition that B(p, w, x̄) is closed, as it is the intersection of two
closed sets: P (x̄) ∩ {x | p · x = w}. Since u is continuous, by Weierstrass theorem it
has a maximum; since u is strictly quasi-concave this maximum is unique (Theorem
A.14): if there were two distinct maxima x∗ ̸= x′∗ , then necessarily u(x∗ ) = u(x′∗ )
and then the mid-point z = (x∗ + x′∗ )/2 is also in B(p, w, x̄) and by strict quasi-
concavity u(z) > u(x∗ ) = u(x′∗ ) − a contradiction. 2

Remark We have ∀t > 0, f (tp, tw) = f (p, w): f is homogeneous of degree zero with
respect to prices and wealth. We then normalize prices, and choose a numéraire:
p1 = 1.

Given differentiabilty, we can compute where the solution to the consumer’s

problem is.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 8

Proposition 1.2 The bundle x∗ is solution of the optimization program of the con-
sumer if and only if it satisfies the first order conditions: there exists λ∗ > 0 such
that (
∇u(x∗ ) = λ∗ p
(1.2)
p · x∗ = w
where ∇u is the gradient of the utility function:
 
∂u ∂u ∂u
∇u(x) = (x), (x), . . . , (x) .
∂x1 ∂x2 ∂xL
Proof: These are the necessary first-order conditions of Lagrange’s method to solve
the optimization problem (1.1). See Theorem A.15.
Let us prove that these necessary conditions are sufficient. We know (Proposition
1.1) that the problem has a solution, and it is unique. So it must be the one satisfying
(1.2).
A more direct proof uses the following property of strictly quasiconcave functions:

∀x, x′ , u(x′ ) ≥ u(x) ⇒ ∇u(x)(x′ − x) ≥ 0.

Now suppose that (x∗ , λ∗ ) satisfies (1.2) but is not a solution of the problem: there
exists x such that p · x = w and u(x) > u(x∗ ). Then for t ∈ [0, 1] sufficiently close
to 1, by continuity of u: p · (tx) < w and u(tx) > u(x∗ ). The above property then
gives: ∇u(x∗ )(tx − x∗ ) ≥ 0; but from (1.2) we get:

∇u(x∗ )(tx − x∗ ) = λ∗ p · (tx − x∗ ) < λ∗ (w − w) = 0,

a contradiction. 2

The geometry of Proposition 1.1 is displayed in Figure 1.1.

Exercise 1.2 Give the expression of the demand function of a consumer whose
utility function is:

1. u(x1 , x2 ) = α1 ln x1 + α2 ln x2 (log-linear).

2. u(x1 , x2 ) = (xr1 + xr2 )1/r (CES).

Remark Consider the expenditure problem:

Min p·x
x ∈ RL (1.3)
s. t. u(x) ≥ ū
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 9

x2
6
w 7

p2 p



budget
line

x∗2

-
O x∗1 w x1
p1

Figure 1.1

Take x∗ solution of (1.1); if ū = u(x∗ ), then x∗ solves (1.3). Conversely, take x∗

solution of (1.3); if w = p · x∗ , then x∗ solves (1.1). We say that (1.1) and (1.3) are
the dual problem of each other.

1.1.3 Offer curve

Consider for simplicity an economy with two goods; for a given bundle of endoments
ω ∈ R2 , and for any market price p, the consumer computes her wealth and opti-
mizes. The wealth is w = p1 · ω = ωa1 + pωa2 ; her optimal choice, x∗ = f (p, p · ω).

Thus, any price p (for good 2, good 1 is the numéraire) posted on the market trig-
gers a demand by the consumer; this defines an offer curve, O(ω), which is the path
followed by the optimal choice x∗ when p varies from 0 to +∞. See Figure 1.2.
The offer curve has two asymptotic directions:

ˆ a horizontal one when p tends toward +∞: then good 1 becomes free, the
assumption of monotonicity leads the agent to demand an infinite quantity of
it;

ˆ a vertical one when p tends to zero: this corresponds to the case when good 2
becomes free.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 10

ω O(ω)
•
Iu(ω)
-
O

Figure 1.2: offer curve

The offer curve O has two additional distinctive features:

ˆ it goes through ω; when the market price is collinear to ∇u(ω), then ω is the
demand of the consumer; she does not find it beneficial to trade;

ˆ it lies above the indifference curve Iu(ω) going through ω; the consumer can
always demand ω and secure a utility level of u(ω); if she demands another
bundle x ̸= ω, it can only be that x brings a higher utility: u(x) > u(ω).

In more rigorous terms: ω ∈ graph O and graph O ⊆ P (ω).

Canonical example: the log-linear utility

Consider a consumer with log-linear utility function and endowment:

u(x1 , x2 ) = α ln x1 + (1 − α) ln x2 and ω = (ω1 , ω2 ) ≫ 0,

with α ∈]0, 1[. The demand of the consumer is (see Exercise 1.2):
 
1−α
x(p) = α(ω1 + pω2 ), (ω1 + pω2 )
p
which are the parametric equations of the offer curve O. We have:

lim x(p) = (+∞, (1 − α)ω2 ) and lim x(p) = (αω1 , +∞).

p→+∞ p→0
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 11

O is a hyperbola with asymptotes:

x1 = αω1 and x2 = (1 − α)ω2 .

Indeed, with the change of variable z1 = x1 − αω1 and z2 = x2 − (1 − α)ω2 , the

equations of O become:
c z1
z2 = and p =
z1 αω2
with c = αω1 (1 − α)ω2 .
One checks that ω ∈ graph O: if x1 = ω1 then x2 = ω2 ; this happens for the
1−α ω1
price p = α ω2
which is the MRS (of good 1 w.r.t. good 2) at ω.
One also checks that graph O ⊆ P (ω); the indifference curve Iu(ω) is hyperbolic
with equation: α
  α−1
x2 x1
= ,
ω2 ω1
with the abscissa and ordinate axis as asymptotes. 2

The notion of offer curve can be extended to any number of goods (it is an
hypersurface).

1.2 Pure exchange

For the sake of simplicity, the analysis is done in the case L = I = 2, and displayed
in the Edgeworth box.

1.2.1 The Edgeworth box

It is drawn on Figure 1.3.
Consider two agents, a et b, and two goods, 1 et 2. Agent a is initially endowed
with (ωa1 , ωa2 ) and agent b with (ωb1 , ωb2 ). In the system of axis (Oa , xa1 , xa2 ), the
coordinates of the origin Ob are (ωa1 + ωb1 , ωa2 + ωb2 ), the axis (Ob , xb1 ) is horizontal
and oriented leftward, and the axis (Ob , xb2 ) is vertical and oriented downward. The
point ω reads ωa in the system of axis (Oa , xa1 , xa2 ) and ωb in the system of axis
(Ob , xb1 , xb2 ). The budget constraint of agent a in the system of axis (Oa , xa1 , xa2 )
and that of agent b in the system of axis (Ob , xb1 , xb2 ) coincide.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 12

xa2 Ob
 6 ωb1
xb1

budget
line p
p2
1

ω
ωa2 • ωb2

xa1
-
Oa ωa1
x
?
b2

Figure 1.3: The Edgeworth box

1.2.2 Walrasian equilibrium

In Figure 1.4, p cannot be an equilibrium price at which exchange takes place.
Indeed, at that price, agent a wants to sell a strictly positive quantity of his initial
endowment in good 1 (net supply) which does not match the quantity of good 1 that
agent b wants to buy (net demand). There even are prices at which both agents
want to sell good 1, or at which both want to buy good 1.
Exchange takes place at a price p for which these quantities match:

ωa1 − x∗a1 = x∗b1 − ωb1

i.e., when the overall demand in good 1, x∗a1 +x∗b1 , equals the overall supply, ωa1 +ωb1 .
One then observes that the same condition will automatically be fulfilled for good
2. Indeed, given that p · x∗a = p · ωa and p · x∗b = p · ωb , one gets:

p1 (x∗a1 + x∗b1 ) + (x∗a2 + x∗b2 ) = p1 (ωa1 + ωb1 ) + (ωa2 + ωb2 ) ,

hence the conclusion.

In the Edgeworth box, the equilibrium occurs when the optimal choices of both
consumers coincide. See Figure 1.5.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 13

net demand of agent b

Ob
6 


x∗a (p, wa )
•

x∗b (p, wb )•
ω
•

-
Oa 
?
net supply of agent a
Figure 1.4: Disequilibrium

Definition 1.1 A walrasian equilibrium is a vector (p∗ , x∗a , x∗b ) such that

 x∗a = fa (p∗ , p∗ · ωa )


x∗b = fb (p∗ , p∗ · ωb ) (1.4)

 x∗ + x∗ = ω + ω

a b a b

Theorem 1.2 [Existence Theorem] There exists a walrasian equilibrium.

One can give a heuristic proof in the Edgeworth box. Both offer curves cross at ω,
where Ob (ω) goes above Oa (ω); but one knows that Oa (ω) has an upward vertical
asymptotic direction and Ob (ω) has a leftward horizontal asymptotic direction; by
continuity and the theorem of intermediate values, they must cross each other at
another point (x∗ on Figure 1.6) than ω. The corresponding equilibrium budget line
is (ω, x∗ ), to which the equilibrium price p∗ is orthogonal.

Exercise 1.3 Compute the walrasian equilibrium of the two consumers–two goods
economy, with utility functions and initial endowments:
ua (xa1 , xa2 ) = α ln xa1 + (1 − α) ln xa2 and ωa = (1, 0) ;
ub (xb1 , xb2 ) = β ln xb1 + (1 − β) ln xb2 and ωb = (0, 1) .
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 14

Ob
 6

x∗a (p, wa )
•
x∗b (p, wb ) ω
•

-
Oa ?

Figure 1.5: Equilibrium

Give the parametric expression of the offer curves.

Remark There can be several equilibria. But there is almost always (except in
non-generic cases) an odd number of equilibria as Figure 1.7 depicts.

Exercise 1.4 Consider a pure exchange economy with two agents characterized by
the utility functions (said to be quasi-linear):
1 1
ua (xa1 , xa2 ) = xa1 − (xa2 )−8 , ub (xb1 , xb2 ) = − (xb1 )−8 + xb2 ,
8 8
and initial endowments ωa = (2, r) and ωb = (r, 2), where r = 28/9 − 21/9 .

1. Show that the offer curves of the two agents are respectively (good 2 is the
numeraire, p is the price of good 1):

OCa (p) = 2 + rp−1 − p−8/9 , p1/9

and
OBb (p) = p−1/9 , 2 + rp − p8/9

Note that the demand of consumer a in good 1 can be increasing in p.

2. Compute the equilibrium price(s) of the economy.

CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 15

Ob
 6

p∗ 

x∗
•

Ob (ω) ω Oa (ω)
•

-
Oa ?

Figure 1.6: Equilibrium and offer curves

(Hints: r has a peculiar form. It should help you find the root of a non-trivial
polynomial expression. For showing that you found all the roots of that expression,
the intermediate value theorem may be useful.)

1.2.3 Intertemporal choice

One can interpret the current problem as that of the intertemporal allocation of
resources; there are L dates and only one good (say, money); xℓ is the consumption
of money at date ℓ.
Consider for example L = 2 and let r be the interest rate, then the consumption
at date 2 is x2 = ω2 + (1 + r)(ω1 − x1 ), giving the budget constraint in terms of
present values:
1 1
x1 + x2 = ω1 + ω2 ,
1+r 1+r
with p1 = 1 being the ‘numéraire’, p2 = 1/(1 + r), and w = ω1 + ω2 /(1 + r).

Exercise 1.4: Compute the walrasian equilibrium of the two consumers, one good
and two dates intertemporal economy, with utility functions ui (xi1 , xi2 ) = ln xi1 +
δ ln xi2 for i = a, b, and arbitrary initial endowments.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 16

Ob
 6

Ob (ω) ω Oa (ω)
•

-
Oa ?

Figure 1.7: Multiple equilibria

1.3 Producers
There are J firms, indexed from 1 to J, where J = {1, . . . , J}. Each firm faces
different possible ways to transform inputs into outputs.

1.3.1 Production sets

For a generic firm, these ways are represented by a production possibility set Y ⊂ RL .
The production set Y has a boundary, denoted ∂Y, described by a map g : RL → R
such that
Y = { y ∈ RL | g(y) ≤ 0 }.

Element y ∈ Y is referred to as a production plan. Negative coordinates represent

inputs and positive outputs. The map g is assumed to be convex and differentiable.
A simple illustration of such a production set will be employed in the case of
two goods, where Y ⊂ R2 is defined by g(y1 , y2 ) = ∥(y1 , y2 )∥ − 1, and ∥(y1 , y2 )∥ =
y12 + y22 is the Euclidean norm on R2 ; then
p

∂Y = {y = (y1 , y2 ) ∈ R2 | ∥(y1 , y2 )∥ = 1}.

The set Y is here the disk of radius 1 centered at O, and its frontier ∂Y the circle
of radius 1; they are represented in Figure 1.8.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 17

Ty ∂Y
y2
6
∂Y

∇ 
*


•

y
Y

-
y1

Figure 1.8: Production

1.3.2 Profit maximization

In some instances, the producer determines its optimal choice of production through
profit maximization. It then solves the following program:



 Max p·y
y ∈ RL

y∈Y

 subject to

It is clear from the assumptions that an optimal choice will be on the frontier ∂Y .

Exercise 1.5 Give the expression, as a function of prices, of the optimal choice
of a producer using goods 2 and 3 to produce good 1, with production function:
√ √
g(y1 , y2 , y3 ) = y1 − y2 − y3 .

1.4 The general model

For j ∈ J , element yj ∈ Yj is referred to as a production plan for firm j. Let
Q
y = (y1 , . . . , yJ ) ∈ Y = j Yj be a list of production plans.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 18

For i ∈ I, consumer i has a utility function ui : RL → R over the space of

consumption bundles. Let us denote xi ∈ RL a consumption bundle for consumer i,
and x = (x1 , . . . , xI ) ∈ RLI a list of consumption bundles.

1.4.1 Endowments and allocations

The consumer is initially endowed with (a vector of) goods x̄i = (x̄i1 , . . . , x̄iL ) ∈ RL
and with an initial portfolio of shares in firm θ̄i = (θ̄i1 , . . . , θ̄iJ ) ∈ RJ , where θ̄ij
represents the share held by consumer i in firm j.
P
Exactly 100% of the shares are initially distributed, hence i θ̄ij = 1 for every
j. The economy is therefore described using the following data:

((ui , x̄i , θ̄i )i∈I , (gj )j∈J ).

An allocation is a feasible way to distribute the available resources between pro-

ducers and consumers.

Definition 1.2 An allocation (x, y) is a list of bundles and plans such that y ∈ Y
and aggregate supply and demand balance each other out:
X X X
xi = x̄i + yj . (1.5)
i i j

1.4.2 Returns on portfolios

If firm j produces the plan yj (a vector in RL , with as many coordinates as goods),
and if consumer i holds a share θij of firm j (a scalar in [0, 1], for example θij = 0.15
if consumer i holds 15% of firm j), then consumer i has a claim on the bundle:
 
yj1
 . 
θij yj = θij  . 
 . 
yjL

(e.g. on 15% of the production plan of firm j).

Altogether, the portfolio θi (a vector in RJ , with as many coordinates as firms2 ),
2
If there are two firms, and consumer i holds 15% of firm 1 and 32% of firm firm 2, her portfolio
is:  
0.15
θi =
0.32
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 19

summing over all firms, yields the following return:

J
X
θij yj . (1.6)
j=1

Given the market price vector p in RL , the market value of this return is (taking
the inner product with p):
J
X J
X L
X
p· θij yj = θij pℓ yjℓ , (1.7)
j=1 j=1 ℓ=1
PL
which is the dividends of firms j (share θij of the profit p·yj = ℓ=1 pℓ yjℓ ), summed
over all firms.
P
The expression (1.6) and (1.7) are cumbersome, as they involve multiple ’s. It
is more parsimonious, in terms of notation, and also for computer programming, to
use matricial forms.
Denote Y the L×J matrix of production plans (its J column vectors are the
production plans, hence its entry at row ℓ and column j is the output - or input if
it is negative - of firm j in good ℓ). Check that we have:
J
X
θij yj = Y θi (1.8)
j=1

and that
J
X L
X
θij pℓ yjℓ = p · Y θi (1.9)
j=1 ℓ=1

1.4.3 Equilibrium
Let Y = (y1 . . . yJ ) denote the L×J matrix of production plans. The initial resources
of consumer i are the addition of her endowments in goods plus the product of her
endowment in shares: x̄i + Y θ̄i . For a market price vector p ∈ RL+ , the budget set of
consumer i is therefore

Bi (p, y) = xi ∈ RL | p · xi ≤ p · (x̄i + Y θ̄i ) .



The notion of (Walrasian) equilibrium stems naturally from this.

Definition 1.3 For plans ȳ ∈ Y, an equilibrium is a vector (p∗ , x∗ , ȳ) where

(x∗ , ȳ) is an allocation, and x∗ is a solution to the consumers’ problems

for every i, x∗i = arg max{ ui (xi ) | xi ∈ Bi (p∗ , ȳ)}.

CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 20

The general equilibrium analysis will only be complete when the choice of pro-
duction plans has been determined endogenously. A crucial point, often discarded
in most textbooks, is to determine whether shareholders or stakeholders agree on
what firms should produce.

1.4.4 Shareholders are unanimous

The form of the budget set Bi (p, y) implies that consumer i evaluates firm j’s plans
based on the value of the shares she holds: θ̄ij p · yj . The greater this value, the bet-
ter. Furthermore, at equilibrium the first-order conditions (1.2) of the optimization
problem for consumers yield:

∇ui (x∗i ) = λ∗i p∗ .

The interpretation is as follows: the relative values of goods as perceived by consumer

i (through the ratio of marginal utilities) and the relative values of goods as priced
by the market equalize. This means that for every i, the normalized (with first
coordinate equal to 1) value vector (or gradient) equals the price vector:
∥
∇ i = p∗ (1.10)

Consumers are fully aligned and therefore they all agree on the best plan, i.e.
that which maximizes the value of production based on market prices. The notion
of general equilibrium follows on from this.

Definition 1.4 A general equilibrium (p∗ , x∗ , y ∗ ) is an equilibrium at which pro-

duction plans of firms maximize profits for the equilibrium prices:

for every j, yj∗ = arg max {p∗ · yj | yj ∈ Yj } .

Theorem 1.3 There exists a general equilibrium.

1.4.5 Value vectors

Agents (both consumers and firms) use value vectors ∇ ∈ RL \{0} to make their
decisions. These vectors are in some cases normalized so that their first coordinates
equals 1. The normalized value vector is denoted as ∇∥ .
A production plan is optimal for a firm provided it maximizes the value of pro-
duction for a certain value vector.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 21

Definition 1.5 Plan yj ∈ Yj is a solution for value vector ∇j if

∇j ·yj ≥ ∇j ·yj′ for all yj′ ∈ Yj .

Here, it is said that ∇j supports yj .

Figure 1.8 represents, for a plan y ∈ ∂Y, the value vector ∇ which supports it:
it is orthogonal to the tangent at y of ∂Y, denoted Ty ∂Y.
The following result is important. It demonstrates that in respect of our hy-
potheses there is a bijective relationship between value vectors and optimal plans.

Lemma 1.1 Assuming Yj is not a singleton, then

ˆ yj ∈ ∂Yj if and only if gj (yj ) = 0;

ˆ yj ∈ Yj is a solution for ∇j ∈ RL \ {0} if and only if yj ∈ ∂Yj and Dgj (yj )

and ∇j are collinear, and point in the same direction;

ˆ for all yj ∈ ∂Yj there is a unique (up to collinearity) value vector in RL \ {0}
supporting yj ; it is denoted as ∇j (yj );

ˆ for all ∇j ∈ RL+ \ {0} there is a unique production plan yj ∈ ∂Yj which is a
solution for ∇j .

Proof: See Appendix ??. 2

1.5 Quadratic utilities

By way of illustration, let us consider the following quadratic utility function:
L  
X 1
ui (xi ) = πℓ γi xiℓ − (xiℓ )2 (1.11)
ℓ=1
2

with π = (π1 , . . . , πL ) ∈ SL+ , where SL is the unit simplex in RL :

n o
SL = x ∈ RL+ |
X
xℓ = 1
ℓ

and SL+ denotes its interior; π is the vector of weightings attributed to the different
goods (independently of i), and γi ∈ R+ is a parameter.
P P
Let ωi = x̄i +Y θ̄i be the total initial resources of consumer i, and Ω = i x̄i + j ȳj
be the vector for the overall resources in the economy. We assume that for every
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 22

i, γi > max{Ωℓ , 1 ≤ ℓ ≤ L} such that the utility function satisfies the required
hypotheses (e.g., monotonicity) on the set of relevant consumptions.
Let ∥ · ∥π denote the π-norm3 on RL . Let 1L (respectively 1J ) be the vector with
L (respectively J) coordinates, all equal to 1. Therefore
1 2 1
ui (xi ) = γ − ∥γi 1L − xi ∥2π .
2 i 2
The indifference (hyper)surfaces of consumer i are (hyper)spheres for the π-distance,
centered on γi 1L , which hence represents the ideal bundle for this consumer, the one
that maximises ui (xi ) and that she would choose if there were no budget constraint.
Figure 1.9 illustrates the configuration in the case of 2 goods and equal weights:
π1 = π2 = 0.5 (Euclidean distance). Indifference curves are arcs of a circle centered
on γi 1L . At the individual optimum x∗i , the indifference curve is tangent to the
budget line, determined by the vector of initial resources x̄i + Y θ̄i and orthogonal
to the price vector p, which bounds the budget set Bi (p, y).
Let Π be the diagonal matrix whose size is L with π on the diagonal, then the
gradient takes the following form:

∇i = ∇ui (xi ) = Π(γi 1L − xi ).

The indifference surfaces being spheres, the gradient points towards the center of the
hypersphere of indifference, i.e. the ideal bundle. (In Figure 1.9, at the individual
∥
optimum x∗i , the normalized gradient ∇i = p points toward the center of the arcs
of circle γi 1L .)
P
The Walrasian equilibrium price vector is easy to calculate. Define Γ = i γi .

Lemma 1.2 At equilibrium, we get

1
p∗ = Π(Γ1L − Ω),
Γ−π·Ω
and there exist (αi∗ )i∈I ∈ SI+ such that for every i,

x∗i = γi 1L − αi∗ (Γ1L − Ω).

3
Let us define the following bilinear form on RL × RL , called the π-inner product: y ⊙π z =
R L
P
ℓ πℓ yℓ zℓ . This defines a norm on (if π has positive coordinates):
v
u L
uX
∥y − z∥π = t πℓ (yℓ − zℓ )2 .
ℓ=1
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 23

x2
γi 1L(=2)
6

*p


•
x̄∗i

Bi (p, y)

•
ωi

-
x1

Figure 1.9: Consumption with quadratic utilities

Proof: The first-order conditions of the individual optimization problem for con-
sumer i gives: ∃λi ∈ R+ such that ∇ui (xi ) = Π(γi 1L − xi ) = λi p. Hence xi =
γi 1L − λi Π−1 p.
P P
Denote Λ = i λi . Adding up the individual consumptions gives us: i xi =
Γ1L − ΛΠ−1 p = Ω, hence:
1 λi
p= Π(Γ1L − Ω) and xi = γi 1L − (Γ1L − Ω).
Λ Λ
The price vector p is normalized, so that p · 1L = 1. As a consequence Λ = Γ − π · Ω.
Hence the expression of p.
Denote αi = λi /Λ, hence the expression of xi .
The equilibrium values of (αi )i∈I ∈ SI+ are determined by the budget equations.
Denote z̃i = γi 1L − ωi the ideal transaction of i and z̃ = i z̃i = Γ1L − Ω the ideal
P

aggregate transaction. Since xi = γi 1L − αi z̃, one gets p · xi = γi − αi p · z̃ (given

the normalization of p). The latter equation associated with the budget equation
p · xi = p · ωi yields
1 1
∀i, αi = (γi − p · ωi ) with p = Πz̃.
p · z̃ π · z̃
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 24

Since γi − p · ωi = p · (γi 1L − ωi ) = p · z̃i , simple transformations yield

p · z̃i z̃ ⊙π z̃i
αi = = for every i,
p · z̃ ∥z̃∥2π

recalling the notation y ⊙π z = ℓ π ℓ y ℓ z ℓ . 2

P

It is easy to verify that at equilibrium the gradients are all collinear:

∇∗i = Dui (x∗i ) = αi∗ Π(Γ1L − Ω).

The results lends itself to a quite simple and intuitive geometrical interpretation.
The bundle γi 1L being the ideal bundle of consumer i, her ideal transaction is
z̃i = γi 1L − ωi ; and the aggregate ideal transaction is z̃ = i z̃i = Γ1L − Ω.
P

The equilibrium bundle x∗i = γi 1L − αi∗ z̃ is therefore the ideal bundle decreased
by a proportion of the ideal aggregate transaction. This proportion is
z̃ ⊙π z̃i
αi∗ = for every i
∥z̃∥2π

(cf. proof of Lemma 1.2). The proportion αi∗ measures the π−projection of the
ideal transaction of i on the ideal aggregate transaction. The less z̃i is collinear to
z̃ (a situation where consumer i’s ideal transaction diversifies the aggregate ideal
transaction) the smaller αi∗ , and thus the larger the equilibrium bundle. In this
model a consumer is rewarded if her transaction diversifies the economy.

In conclusion of this first chapter, we can see that complete alignment between
consumers stems from the hypothesis of a perfect market, thus ridding the decision-
making process in firms of any political content. This is no longer the case when
markets fail. Consumers then disagree about the plan to produce. In our model,
∥
this occurs when the ∇i ’s are different from one agent to another, giving rise to
a problem of a political nature when it comes to taking collective decisions, e.g.
choosing the production plans in firms.

Problem set 1.1: Quadratic utilities

Consider the following quadratic utility function over two goods (1 and 2):
   
1 2 1 2
ui (xi ) = γi xi1 − (xi1 ) + γi xi2 − (xi2 ) (1.12)
2 2
and γi ∈ R+ is a parameter.
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 25

Let Ω = (Ω1 , Ω2 ) be the vector for the overall resources in the economy. We
assume that for every i, γi > max{Ω1 , Ω2 } such that the utility function satisfies
the required hypotheses (e.g., monotonicity) on the set of relevant consumptions.

1. Let ∥ · ∥ denote the Euclidean norm on R2 . Let 12 = (1, 1). Show that
1
ui (xi ) = γi2 − ∥γi 12 − xi ∥2 .
2
What can you infer about the shape of the indifference curves?

2. Show that the demand of consumer i is:

(
xi1 = γi − λi p1
(1.13)
xi2 = γi − λi p2

where λi is the Lagrange multiplier of her optimization problem.

P P
3. Denote Λ = i λi and Γ =
i γi . We normalize prices such that: p1 + p2 = 1.
P
Show that market clearing ( i xi = Ω) yields the following expression of the
equilibrium price vector:
Γ − Ω1

 p1 =

2Γ − Ω1 − Ω2 (1.14)
Γ − Ω2
 p2 =

2Γ − Ω1 − Ω2

4. Check that at equilibrium all gradients are collinear, and that the demand of
consumer i is: (
xi1 = γi − αi (Γ − Ω1 )
(1.15)
xi2 = γi − αi (Γ − Ω2 )
with αi = λi /Λ.

5. We want to find the equilibrium value of αi using the budget equations. The
bundle γi 12 being the ideal bundle of consumer i, define her ideal transaction:

z̃i = γi 12 − ωi

and the ideal aggregate transaction:

z̃i = Γ12 − Ω.
X
z̃ =
i

(a) Show that Equation (1.14) can be rewriten:

1
p= z̃.
12 · z̃
CHAPTER 1. PERFECT MARKETS: POSITIVE ANALYSIS 26

(b) Show that Equation (1.15) can be rewriten:

xi = γi 12 − αi z̃. (1.16)

(c) Show that given with the budget equation p · xi = p · ωi , Equation (1.16)
yields
1
∀i, αi = (γi − p · ωi ).
p · z̃
(d) Show that
p · z̃i z̃ · z̃i
αi = = for every i.
p · z̃ ∥z̃∥2
(e) Explain why the less z̃i is collinear to z̃ (a situation where consumer i’s
ideal transaction diversifies the aggregate ideal transaction) the smaller
αi , and thus the larger the equilibrium bundle. Interpret.
 Part II

Imperfect Financial Markets

27
 Part III

Externalities and Public Goods

28
 Part IV

Market Power

29
 Part V

Information

30
 Part VI

Political Economy of Imperfect

Markets

31
Appendix A

Mathematical appendix

(With the courtesy of Prof. Dr. Carlos Alós-Ferrer.)

A.1 Sets and relations

A set X is a collection of elements; x ∈ X - x is an element in X; A ⊆ X - subset
of X; ∅ - the empty set; 2X , P(X) - the set of all subsets of X; Y ⊆ 2X - family of
sets; (Ai )i∈I - family of sets indexed by i ∈ I

Definition A.1 A partition is a family of sets (Ai )i∈I with

ˆ ∪i∈I Ai = X

ˆ Ai ∩ Aj = ∅ for all i, j ∈ I

The Cartesian product X×Y : { (x, y) | x ∈ X, y ∈ Y }; we denote ×i∈I Xi the

Cartesian product of (Xi )i∈I , { (xi )i∈I | ∀ i ∈ I : xi ∈ Xi }

A relation R ⊆ X×X on X is denoted: xRy ⇔ (x, y) ∈ R.

Some relations:

ˆ ≥, = and > on R

ˆ ≥ on R2 with (x1 , x2 ) ≥ (y1 , y2 ) ⇒ (x1 ≥ y1 and x2 ≥ y2 )

ˆ = and > on R2 with x > y ⇒ (x ≥ y and x ̸= y)

ˆ ⪰ on X with x ⪰ y ⇔ x is at least as good as y (preference)

32
APPENDIX A. MATHEMATICAL APPENDIX 33

ˆ ∼ on X with x ∼ y ⇔ (x ⪰ y and y ⪰ x) (indifference)

ˆ ≻ on X with x ≻ y ⇔ (x ⪰ y and y ̸⪰ x) (strict preference)

Properties of relations R ⊆ X×X:

ˆ reflexive - ∀ x ∈ X : xRx

ˆ symmetric - ∀ x, y ∈ X : xRy ⇒ yRx

ˆ asymmetric - ∀ x, y ∈ X : xRy ⇒ y̸Rx

ˆ antisymmetric - ∀ x, y ∈ X : (xRy and yRx) ⇒ x = y

ˆ complete - ∀ x, y ∈ X : xRy or yRx (R complete ⇒ R reflexive)

ˆ transitive - ∀ x, y, z ∈ X : (xRy and yRz) ⇒ xRz

Examples: ≥ on R: reflexive, complete, antisymmetric, transitive; > on R2 :

asymmetric, antisymmetric, transitive.

Definition A.2 An equivalence relation is a relation that is reflexive, symmetric

and transitive

Lemma A.1 Suppose R is an equivalence relation. Let Ca = { x ∈ X | xRa }

ˆ x ∈ Ca ⇒ Cx = Ca

ˆ x∈
/ Ca ⇒ Cx ∩ Ca = ∅

Proof: Let us first show: x ∈ Ca ⇒ Cx = Ca .

def. def. transitivity
To show Cx ⊆ Ca : y ∈ Cx ⇔ yRx; x ∈ Ca ⇔ xRa; (yRx and xRa) ⇒
def.
yRa ⇔ y ∈ Ca .
def. def. symmetry
To show Ca ⊆ Cx :y ∈ Ca ⇔ yRa; x ∈ Ca ⇔ xRa ⇔ aRx;
transitivity definition
(yRa and aRx) ⇒ yRx ⇔ y ∈ Cx .
Let us now show: x ∈
/ Ca ⇒ Cx ∩ Ca = ∅. Assume x ∈
/ Ca and Cx ∩ Ca ̸= ∅:
def. symmetry transitivity
y ∈ Cx ∩ Ca ⇔ yRx and yRa; yRx ⇔ xRy; (xRy and yRa) ⇒
def.
xRa ⇔ x ∈ Ca , contradicting x ∈
/ Ca .

Theorem A.1 Suppose R is an equivalence relation. Then (Ca )a∈X is a partition.

Suppose (Ai )i∈I is a partition. Then R defined by

xRa ⇔ ∃ i ∈ I : x, a ∈ Ai

is an equivalence relation.
APPENDIX A. MATHEMATICAL APPENDIX 34

Proof: Let us show (Ca )a∈X is a partition:

∀ a ∈ X : a ∈ Ca (reflexivity) ⇒ ∀ x ∈ X ∃ a ∈ X : x ∈ Ca
Lemma A.1 ⇒ ∀ x, a ∈ X : either Cx = Ca or Cx ∩ Ca = ∅

Let us now that show (Ai )i∈I induces an equivalence relation:

(
i = j ⇒ xRy, yRx
∀ x ∈ Ai , y ∈ Aj :
i ̸= j ⇒ x̸Ry, y̸Rx

Therefore R is reflexive, symmetric and transititve. 2

A.2 Topology
Definition A.3 A norm on a vector space V over R is a function ∥ · ∥ : V → R
such that:

ˆ ∀ x ∈ V : ∥x∥ = 0 ⇔ x = 0

ˆ ∀ x ∈ V : ∥x∥ ≥ 0

ˆ ∀ x, y ∈ V : ∥x + y∥ ≤ ∥x∥ + ∥y∥ (triangle inequality)

ˆ ∀ x ∈ V, α ∈ R : ∥αx∥ = |α|∥x∥

Claim A.1 (Rℓ , ∥ · ∥) with the Euclidean norm:

Xℓ
∥x∥ = ( x2i )1/2
i=1

is a normed vector space.

Theorem A.2 (Cauchy-Schwarz) For x, y ∈ Rℓ ,

ℓ
!2 ℓ
! ℓ
!
X X X
xi yi ≤ x2i yi2
i=1 i=1 i=1

Proof: Trivially ∀ λ ∈ R
ℓ
X ℓ
X ℓ
X ℓ
X
0 ≤ (xi − λyi )2 = x2i − 2λ xi y i + λ 2 yi2
i=1 i=1 i=1 i=1
APPENDIX A. MATHEMATICAL APPENDIX 35

yi2 > 0, and λ = yi2 ,

P P P
For y ̸= 0, so i i xi y i / i

x2i − 2 λ xi yi + λ2 yi2
P P P
0 ≤ i i i

( i xi yi ) 2
P P
2 xy
yi2
P P P
Pi i 2 i
= i xi − 2 i xi yi + ( i yi2 )2 i
P
i yi

( i xi yi ) 2 ( i x2i )( i yi2 )−( i xi yi )2

P P P P
2
P
= i xi − =
P 2 P 2
i yi i yi

x2i )( ℓi=1 yi2 ). 2

Pℓ Pℓ
xi yi )2 ≤ (
P
Therefore ( i=1 i=1

Proof of Claim A.1: Cauchy-Schwarz implies: for all x, y ∈ Rℓ ,

Pℓ
i=1 xi yi ≤
[( ℓi=1 x2i )( ℓi=1 yi2 )]1/2 so
P P

∥x + y∥2 = 2
P
i (xi + yi )
2
P P P 2
= i xi + 2 i xi y i + i yi
P 2 P 2 P 2 1/2
yi2
P
≤ i xi + 2 [( i xi )( i yi )] + i
2 2
= ∥x∥ + 2∥x∥ ∥y∥ + ∥y∥
= (∥x∥ + ∥y∥)2

Therefore ∥x + y∥ ≤ ∥x∥ + ∥y∥. 2

The open ball B(x, ε) with center x and radius ε > 0 is defined by

B(x, ε) = { y ∈ X | ∥x − y∥ < ε }

The closed ball Bε [x] with center x and radius ε > 0 is defined by

B[x, ε] = { y ∈ X | ∥x − y∥ ≤ ε }

A ⊆ X is bounded provided: ∃ x ∈ X, ε > 0 : A ⊆ B[x, ε]

Definition A.4 A set U ⊆ X is open if and only if:

∀ x ∈ U ∃ ε > 0 : B(x, ε) ⊆ A

A set C ⊆ X is closed if and only if: C ∁ is open

Proposition A.1 X and ∅ are open and closed in X and

ˆ For (Ui )i∈I , ∀ i ∈ I : Ui is open ⇒ ∪i∈I Ui is open

ˆ For (Ci )i∈I , ∀ i ∈ I : Ci is closed ⇒ ∩i∈I Ci is closed

APPENDIX A. MATHEMATICAL APPENDIX 36

Proof: X is open because: ∀ x ∈ X, ε > 0 : B(x, ε) ⊆ X. And ∅ = X ∁ ⇒ ∅ is

closed.
∅ is open because: ∀ x ∈ ∅, ε > 0 : B(x, ε) ⊆ ∅ (∅ is empty). And X = ∅∁ ⇒ X is
closed

x ∈ ∩i∈I Ui
⇒ ∀ i ∈ I ∃ εi > 0 : B(x, ε) ⊆ Ui
⇒ ∀ ε ≤ mini∈I {εi } : B(x, ε) ⊆ ∩i∈I Ui

(∪i∈I Ci )∁ = ∩i∈I Ci∁ (De Morgan laws). 2

Proposition A.2 For (Ui )i∈I and (Ci )i∈I :

∀ i ∈ I : (Ui is open and I is finite) ⇒ ∩i∈I Ui is open.

∀ i ∈ I : (Ci is closed and I is finite) ⇒ ∪i∈I Ci is closed.

Proof: (∀ x ∈ ∪i∈I Ai ∃ i ∈ I : x ∈ Ai ) ⇒ (∃ ε > 0 : B(x, ε) ⊆ Ai ) ⇒ (B(x, ε) ⊆

∪i∈I Ai ).
(∩i∈I Ci )∁ = ∪i∈I Ci∁ (De Morgan laws). 2

Definition A.5 A function f : X → Y is continuous at x0 if and only if

∀ ε > 0 ∃ δ > 0 ∀ x ∈ X : x ∈ B(x0 , δ) ⇒ f (x) ∈ B(f (x0 ), ε)

f is continuous if and only if f is continuous at all x0 ∈ X.

Theorem A.3 For a function f : X → Y the following statements are equivalent:

(1) f is continuous

(2) ∀ A ⊆ Y : A is open ⇒ f −1 (A) is open

(3) ∀ C ⊆ Y : C is closed ⇒ f −1 (C) is closed

Definition A.6 A family of sets (Ui )i∈I is an open cover of A ⊆ X if and only if
Ui is open for all i ∈ I and A ⊆ ∪i∈I Ui

Definition A.7 A subset A ⊆ X is compact if and only if for all open covers
(Ui )i∈I there is a finite open subcover (Ui )i∈J with J finite

Observation A.1 If A is compact, then A is bounded

APPENDIX A. MATHEMATICAL APPENDIX 37

Corollary A.1 (Heine-Borel) A ∈ Rℓ is compact if and only if A closed and

bounded

Theorem A.4 A continuous function f : X → Y , if C ⊆ X is compact, then

f (C) ⊆ Y is compact

Theorem A.5 (Weierstrass) If A is compact and F is continuous, then arg maxA f

is non-empty and closed

u(S) = { a ∈ R | ∃ x ∈ S :
Th. A.4
Proof: arg maxS u is non-empty: S is compact ⇒
Heine-Borel
u(x) = a } is compact ⇒ u(S) is closed and bounded ⇒ sup u(S) exists and
u(S) is closed ⇒ sup u(S) ∈ u(S) ⇒ maxS u = sup u(S)
Th A.3
arg maxS u is closed maxS u is closed ⇒ u−1 (max u(S)) is closed. 2

A.3 Linear algebra

In the sequel, V is a real vector space. A vector v ∈ V is a linear combination of
v1 , . . . , vn if there exist scalars α1 , . . . , αn such that

v = α1 v1 + . . . + αn vn

Definition A.8 A subset X of V spans V if every vector in V can be written as

a linear combination of a finite number of elements of X.

Example: X = {(1, 0), (0, 1)} spans R2 .

Definition A.9 A finite set of vectors (v1 , . . . , vn ) is linearly dependent if there

exist scalars α1 , . . . , αn with at least one αi ̸= 0 such that

α1 v1 + . . . + αn vn = 0

A finite set of vectors (v1 , . . . , vn ) is linearly independent if it is not linearly

dependent. An infinite set of vectors is linearly independent if each finite subset is
linearly independent.

Definition A.10 A basis of V is a set of linearly independent vectors that spans

v.
APPENDIX A. MATHEMATICAL APPENDIX 38

Example: The canonical basis of Rn is the set of vectors {e1 , . . . , en }, where ei =

(0, . . . , 0, 1, 0, . . . , 0) with 1 in the ith position: eii = 1 and eji = 0 for j ̸= i.

Theorem A.6 Let B = {bi ∈ V, i ∈ I} be a basis for V . Then every non-zero

vector v ∈ V has a unique representation as a linear combination, with coefficients
not all zero, of a finite number of vectors in B.
P
Proof: Let v ̸= 0. Since B spans V : v = i∈I1 αi bi , I1 ⊆ I, I1 finite.
P
Suppose: v = i∈I2 βi bi , I2 ⊆ I, I2 finite.
Let I3 = I1 ∪ I2 and let αi = 0 for i ∈ I3 \I1 and βi = 0 for i ∈ I3 \I2 . Then
X X
v= αi bi = β i bi
i∈I3 i∈I3

− βi )bi = 0. Hence αi = βi for all i. 2

P
As a consequence i∈I3 (αi

Any two bases of a vector space have the same cardinality, called dimension.

Theorem A.7 In a vector space of finite dimension, all bases have the same number
of elements. It is the dimension of v: dim V .

Proof: Suppose to the contrary that U = {u1 , . . . , un } and W = {w1 , . . . , wm } are

two bases such that m > n.
Since U spans V : w1 = ni=1 α1i ui , with one α1i ̸= 0; w.l.o.g. let α11 ̸= 0. Hence
P

u1 is a linear combination of w1 , u2 . . . , un .
Pn
Since U spans V , these vectors also span V , hence w2 = α21 w1 + i=2 α2i ui ,
with one α2i ̸= 0 for i = 2, . . . , n; w.l.o.g. let α22 ̸= 0.
Iterating the argument: {w1 , . . . , wn } spans V ; hence wn+1 is a linear combina-
tion of these vectors - a contradiction. 2

Definition A.11 Let V and W be two real vector spaces. A mapping f : V → W

is a linear mapping if, for all α, β ∈ K and all u, v ∈ V ,

f (αu + βv) = αf (u) + βf (v)

Examples:

ˆ The identity 1V : V → V .

ˆ The null mapping 0V (v) = 0 for all v ∈ v.

APPENDIX A. MATHEMATICAL APPENDIX 39

ˆ The projections pi : Rn → R with pi (v 1 , . . . , v n ) = v i .

Proposition A.3 Let f : V → W be a linear mapping.

1. f (0) = 0.

2. f (−v) = −f (v) for all v ∈ V .

3. If U is a subspace of V , f (U ) is a subspace of W .

4. If U ′ is a subspace of W , f −1 (U ′ ) is a subspace of V .

5. If (v1 , . . . , vn ) are linearly dependent, then (f (v1 ), . . . , f (vn )) are linearly de-
pendent.

6. If X ⊆ V spans V , then f (X) spans f (V ).

7. If g : U → V is a linear mapping, then f ◦ g : U → W is a linear mapping.

Let f : V → W be a linear mapping. Let Imf denote its image (or range), and let
Kerf denote its kernel:

Kerf = f −1 (0) = {v ∈ V | f (v) = 0}

Proposition A.4 A linear mapping f : V → W is injective if and only if Kerf = 0.

Proof: If f is injective, since f (0) = 0, necessarily Kerf = 0.

Suppose Kerf = 0; let u, v ∈ V such that f (u) = f (v). Then f (u − v) = 0, i.e.
u − v ∈ Kerf , hence u = v. 2

An isomorphism is a bijective linear mapping. Then V and W are said isomorphic,

denoted V ∼
= W.

Proposition A.5 The inverse of an isomorphism is also an isomorphism.

Proof: f : V → W bijective; f −1 : W → V its inverse (also bijective). Let us prove

it is linear.
Let α1 , α2 ∈ K, and w1 , w2 ∈ W . Consider v = α1 f −1 (w1 ) + α2 f −1 (w2 ).
Apply f on both sides: f (v) = α1 f (f −1 (w1 )) + α2 f (f −1 (w2 )) = α1 w1 + α2 w2 .
Apply f −1 on both sides: f −1 (α1 w1 + α2 w2 ) = v, hence f −1 is linear. 2

Proposition A.6 Let f : V → W be a linear mapping and V be finite dimensional,

then
dim V = dim Kerf + dim Imf
APPENDIX A. MATHEMATICAL APPENDIX 40

A.4 Matrices
Definition A.12 An m × n real matrix is a m-tuple of vectors of Rn written as
 
a a12 . . . a1n
 11 
 a21 a22 . . . a2n 
A= .
 
 .. .. .. .. 
 . . . 

am1 am2 . . . amn

The set of m × n real matrices is denoted Rmn .

It displays m row vectors of Rn , or n column vectors of Rm .

The transposed matrix of A = [aij ] is the n × m matrix At = [atij ] where atij = aji .

Sum and product by a scalar: A + B = [aij + bij ] and λA = [λaij ].

Definition A.13 Let A be m × n and B be n × p. The product matrix c = AB is

the m × p matrix with generic element
n
X
cij = aik bkj
k=1

Example:  
1 0
!  !
−1 2 0 1 1
 1 
= 5 0
3 −2 1 0  −3
 3 
 −2 1
4 −2

Proposition A.7 Let A ∈ Rmn , B, C ∈ Rnp , D ∈ Rpq .

ˆ (AB)t = B t At

ˆ A(B + C) = AB + AC and (B + C)D = BD + CD

ˆ In general, AB ̸= BA (even if m = n = p).

In the sequel, we consider the canonical base for each real vector space.

Theorem A.8 Let V, W be real vector spaces with dim V = n, dim W = m.

1. For any matrix A ∈ Rmn , the mapping fA : V → W such that for all v ∈ V
fA (v) = Av is linear.
APPENDIX A. MATHEMATICAL APPENDIX 41

2. For any linear mapping f : V → W , there exists a unique matrix Af ∈ Rmn

such that f (v) = Af v for all v ∈ V .

Proposition A.8 Let U, V, W be finite-dimensional real vector spaces, and f : V →

W , g : U → V be linear mappings. Let A, B be the matrices of f and g with respect
to some fixed bases. Then AB is the matrix of f ◦ g : U → W .

Definition A.14 Let f : V → W be a linear mapping, V, W finite-dimensional real

vector spaces. The rank of f is:

rank f = dim Imf (= dim V − dim Kerf )

Proposition A.9 Let f : V → W be a linear mapping; dim V = n and dim W = m.

Let A be the m × n matrix of f . Then rank f is equal to:

ˆ the maximum number of linearly independent column vectors in A;

ˆ the maximum number of linearly independent row vectors in A.

Hence rank f ≤ m and rank f ≤ n. Moreover

ˆ f is injective if and only if rank f = dim V ;

ˆ f is surjective if and only if rank f = dim W .

Definition A.15 A square matrix A ∈ Rn is invertible if there exists A−1 ∈ Rn

2 2

such that  
1 0 ... 0
 
 0 1 ... 0 
A−1 A = A A−1 = In = 
 
.. .. .. .. 

 . . . . 

0 0 ... 1

Proposition A.10 Given a square matrix A ∈ Rn :

2

ˆ A is invertible if and only if it has full rank.

ˆ For all A, B ∈ Rn , (AB)−1 = B −1 A−1 .

2

Definition A.16 Let V be a real vector space. A function f : V × V → R is a

bilinear form if it is linear in both arguments, i.e. for α, β ∈ K and u, v, w ∈ V ,
APPENDIX A. MATHEMATICAL APPENDIX 42

1. f (αu + βv, w) = αf (u, w) + βf (v, w) and

2. f (w, αu + βv) = αf (w, u) + βf (w, v)

It is symmetric if f (u, v) = f (v, u) for all u, v ∈ V .

Example: Let V be n-dimensional and fix a basis. The inner product (or scalar
product) is
n
X
u·v = ui vi
i=1

Proposition A.11 Let V be a real vector space with dim V = n. In the canonical
basis:

1. ∀A ∈ Rn , the mapping fA : V × V → K given by fA (u, v) = ut Av for all

2

u, v ∈ V is a bilinear form;

2. for any bilinear form f : V × V → R, there exists a unique matrix Af ∈ Rn

2

such that f (u, v) = ut Af v for all u, v ∈ V ;

3. f is symmetric if and only if Af is symmetric.

Proof: 1/ and 3/ are obvious. For 2/,let A = [aij ] with aij = f (ei , ej ). For u =
(αi )ni=1 and v = (βj )nj=1 (in B), bilinearity implies f (u, v) = ni,j=1 αi βj aij = ut Av.
P

Definition A.17 A quadratic form is a mapping q : V → R such that there exists

a symmetric bilinear form f on V such that q(v) = f (v, v) for all v ∈ V .

Example: The inner product is clearly symmetric. Therefore the mapping q(x) =
x · x = ni=1 x2i is a quadratic form.
P

As a consequence of Proposition A.11 quadratic forms and symmetric matrices are

equivalent:
n
X n X
X n
t
q(x) = x Ax = aii x2i +2 aijxi xj
i=1 i=1 j=i+1

Definition A.18 A quadratic form q : V → R is

ˆ positive semidefinite if q(v) ≥ 0 for all v ∈ V ;

ˆ positive definite if q(v) > 0 for all v ∈ V \{0};

APPENDIX A. MATHEMATICAL APPENDIX 43

ˆ negative semidefinite if q(v) ≤ 0 for all v ∈ V ;

ˆ negative definite if q(v) < 0 for all v ∈ V \{0}.

Example: the quadratic form associated to a scalar product is positive definite.

Observation A.2 Let A ∈ Rn be the matrix of a quadratic form; if it is diagonal

2

ˆ it is positive (resp. negative) semidefinite if aii ≥ 0 (resp. ≤ 0) for all i;

ˆ positive (resp. negative) definite if if aii > 0 (resp. < 0) for all i;

A.5 Optimization
Consider the unconstrained problem

Maximize f (x)
(A.1)
subject to x ∈ X
where the objective function f : X ⊆ Rn → R is real-valued.

Definition A.19 Let f : X ⊆ Rn → R be a real-valued function and x∗ ∈ X.

ˆ x∗ is a local maximum of f if ∃ϵ > 0 : f (x∗ ) ≥ f (x) ∀x ∈ B(x∗ , ϵ) ∩ X

ˆ x∗ is a global maximum of f if f (x∗ ) ≥ f (x) ∀x ∈ X

Definition A.20 Let f : X ⊆ Rn → Rm be differentiable at x̄ ∈ IntX. Then all

partial derivatives
∂fi
(x̄), i = 1, . . . , m, j = 1, . . . , n
∂xj
exist. Moreover the matrix of Df (x̄) with respect to the canonical basis is given by
the m × n Jacobian matrix
 
∂f1 ∂f1 ∂f1
∂x1 ∂x2
... ∂xn
 
∂f2 ∂f2 ∂f2

∂x1 ∂x2
... ∂xn

Jf (x̄) = 
 
.. .. .. .. 

 . . . . 

∂fm ∂fm ∂fm
∂x1 ∂x2
... ∂xn

Let f : X ⊆ Rn → R be real-valued (m = 1), the Jacobian is a row vector called

gradient:  
∂f ∂f ∂f
∇f (x̄) = (x̄) (x̄) . . . (x̄)
∂x1 ∂x2 ∂xn
Let us give the first-order necessary conditions (FONC).
APPENDIX A. MATHEMATICAL APPENDIX 44

Theorem A.9 Let f : X ⊆ Rn → R; if x∗ ∈ IntX, f is differentiable at x∗ and x∗

is a local maximum of f , then
∇f (x∗ ) = 0

Proof: Let us show that the partial derivatives at x∗ are 0. Let ei be the ith vector
in the canonical basis. Define gi (h) = f (x∗ + hei ). By construction, 0 is a local
maximum (say) of gi hence
1 ∂f ∗
0 = gi′ (0) = lim (f (x∗ + hei ) − f (x∗ )) = (x )
h→0 h ∂xi
The points at which ∇f (x∗ ) = 0 are the critical points. 2

Definition A.21 Let f : X ⊆ Rn → R be a real-valued function twice differentiable

at x̄ ∈ IntX. Then D2 f (x̄) is a bilinear form. The matrix associated to D2 f (x̄)
with respect to the canonical basis is the Hessian matrix
 2 
∂ f1 ∂ 2 f1 2f

∂ 2 x1 ∂x1 ∂x2
. . . ∂x∂1 ∂x 1
n
 2 
 ∂ f2 ∂ 2 f2 ∂ 2 f2
∂ 2 x2
. . . ∂x2 ∂xn 
Hf (x̄) =  ∂x2.∂x1
 
.. .
.. .. .. 

 . . 

2
∂ fm 2
∂ fm 2

∂xn ∂x1 ∂xn ∂x2

. . . ∂∂ 2fxmn

Theorem A.10 (Schwartz theorem) Let f : X ⊆ Rn → R be twice continuously

differentiable in a neighborhood of x̄ ∈ IntX, then the Hessian matrix Hf (x̄) is
symmetric.

Second order sufficient conditions (SOSC).

Theorem A.11 Let f : X ⊆ Rn → R and x∗ ∈ IntX. Suppose f is twice continu-

ously differentiable in a neighborhood of x∗ .

1. If ∇f (x∗ ) = 0 and Hf (x∗ ) is negative definite, then x∗ is a strict local maxi-

mum of f .

2. If x∗ is a local maximum of f , then Hf (x∗ ) is negative semidefinite.

Proof: 1/ ∃ϵ > 0 such that f is C 2 on B(x∗ , ϵ) and Hf (x) is negative definite

∀x ∈ B(x∗ , ϵ). By Taylor’s Theorem ∀x ∈ B(x∗ , ϵ) ∃c ∈ B(x∗ , ϵ):

f (x) − f (x∗ ) = ∇f (x∗ ) · (x − x∗ ) + 21 (x − x∗ )t Hf (c)(x − x∗ )

= 12 (x − x∗ )t Hf (c)(x − x∗ ) < 0 (as ∇f (x∗ ) = 0)

APPENDIX A. MATHEMATICAL APPENDIX 45

2/ For n = 1: suppose to the contrary that f ′′ (x∗ ) > 0 hence by 2/ x∗ is a strict

local minimum, a contradiction.
For n > 1: define g(h) = f (x∗ + hx) for h ∈ R and a fixed x; g(0) = f (x∗ ) and
by construction 0 is a local maximum of g, hence g ′′ (0) ≤ 0 (case n = 1).
Chain Rule yields g ′ (h) = ∇f (x∗ + hx) x and g ′′ (h) = xt Hf (x∗ + hx) x. Hence
g ′′ (0) = xt Hf (x∗ ) x ≤ 0. 2

Remark: Minima of f are maxima of −f .

Definition A.22 Let f : X ⊆ Rn → R, where X is convex. If for all x, y ∈ X and

t ∈ [0, 1]

ˆ f (tx + (1 − t)y) ≥ tf (x) + (1 − t)f (y), f is concave;

ˆ f (tx + (1 − t)y) ≥ min{f (x), f (y)}, f is quasiconcave.

Those definitions are strict when the corresponding inequality is strict whenever
x ̸= y and t ∈]0, 1[.

The function f is (quasi)convex if (−f ) is (quasi)concave

Proposition A.12 Every concave function is quasiconcave.

The upper contour set: Uf (x) = {y ∈ X|f (y) ≥ f (x)}.

Proposition A.13 f is (strictly) quasiconcave if and only if Uf (x) is (strictly)

convex for all x ∈ X.

Proof: (⇒) Take t ∈ [0, 1], and y, z ∈ Uf (x): both f (y), f (z) ≥ f (x), hence
min{f (y), f (z)} ≥ f (x) and by quasiconcavity f (ty + (1 − t)z) ≥ f (x), hence
tx + (1 − t)y ∈ Uf (x).

(⇐) Take t ∈ [0, 1], x, y ∈ X. If f (y) ≥ f (x), then x, y ∈ Uf (x); by convexity

tx + (1 − t)y ∈ Uf (x), hence f (tx + (1 − t)y) ≥ f (x) = min{f (x), f (y)}. 2

Theorem A.12 Let f : X ⊆ Rn → R be C 2 , and X open and convex.

1. f is concave on X if and only if Hf (x) is negative semidefinite for all x ∈ X;

2. if Hf (x) is negative definite for all x ∈ X, then f is strictly concave on X.

APPENDIX A. MATHEMATICAL APPENDIX 46

Beware: Reverse of 2/ does not hold; for example f (x) = −x4 is strictly concave
but Hf (0) is not negative definite.
For concave functions, the FONC of a maximization problem are sufficient.

Theorem A.13 Let f : X ⊆ Rn → R be concave and differentiable, X open and

convex. Then x∗ is a global maximum if and only if ∇f (x∗ ) = 0.

Proof: (⇐) follows from Th. A.12 and the following lemma. 2

Lemma A.2 Let f : X ⊆ Rn → R be differentiable, X open and convex. Then f is

concave if and only if

∇f (x)(y − x) ≥ f (y) − f (x) ∀x, y ∈ X

Proof: Obvious is f is C 2 thanks to Taylor’s Theorem:

f (y) − f (x) = ∇f (x) · (y − x) + 1/2(y − x)t Hf (c)(y − x)

| {z }
≤0

Theorem A.14 Let f : X ⊆ Rn → R be strictly quasiconcave, and X convex. Then

1. any local maximum of f is global;

2. if there is such a maximum, it is unique and strict.

Proof: 1/ Let x∗ be a local maximum. Suppose ∃y ∈ X: f (y) > f (x∗ ). By

∗ ∗ ∗
strict quasiconcavity f (tx + (1 − t)y) > min{f (x ), f (y)} = f (x ) ∀t ∈]0, 1[. But
∥tx∗ + (1 − t)y − x∗ ∥ = |1 − t|∥y − x∗ ∥, a contradiction to x∗ being a local maximum
when t → 1.

2/ Let x ̸= y be two global maxima; in particular f (x) = f (y). Then ∀t ∈]0, 1[

f (tx + (1 − t)y) > f (x), a contrad. 2

Consider now the constrained problem

Optimize f (x)
(A.2)
subject to g(x) = 0
where the objective function f : X ⊆ Rn → R is real-valued and the constraint
function g : X ⊆ Rn → Rm imposes m constraints.

Define the (auxiliary) Lagrangean function

m
X
L(x, λ) = f (x) + λi gj (x) [= f (x) + λ · g(x)]
j=1

where the variables λ = (λ1 , . . . , λm ) are called Lagrange’s multiplicators.

APPENDIX A. MATHEMATICAL APPENDIX 47

Theorem A.15 (Lagrange’s first order conditions) Consider problem (A.2) where
f and g are C 1 . If

1. x∗ ∈ IntX is a local solution of (A.2) and

2. rank Jg (x∗ ) = m

then there exists λ∗ ∈ Rm such that

∂L ∗ ∗
∀i = 1, . . . , n, (x , λ ) = 0
∂xi
Proof: Jg (x∗ ) is m × n, hence 2/ ⇒ m ≤ n. There exists an invertible m × m minor,
suppose w.l.o.g. it corresponds to the m first variables. Denote y = (x1 , . . . , xm )
and z = (xm+1 , . . . , xn ).
By Implicit Function Theorem: ∃h diff. and δ > 0 s.t. ∀x ∈ B(x∗ , δ), g(x) =
0 ⇔ y = h(z).
Differentiating g(h(z), z) = 0:

Jg,y (h(z), z) · Jh (z) + Jg,z (h(z), z) = 0

and
Jh (z ∗ ) = −[Jg,y (y ∗ , z ∗ )]−1 · Jg,z (y ∗ , z ∗ )

Consider now F (z) = f (h(z), z); z ∗ is a local (unconstrained) optimum of F ,

hence 0 = ∇F (z ∗ ) = ∇fy (h(z ∗ ), z ∗ ) · Jh (z ∗ ) + ∇fz (h(z ∗ ), z ∗ ), then

−∇fy (x∗ ) · [Jg,y (x∗ )]−1 · Jg,z (x∗ ) + ∇fz (x∗ ) = 0

Define λ∗ = −∇fy (x∗ )·[Jg,y (x∗ )]−1 , then ∇fz (x∗ )+λ∗ ·Jg,z (x∗ ) = 0 and by definition
of λ∗ : ∇fy (x∗ ) + λ∗ · Jg,y (x∗ ) = 0 hence ∇f (x∗ ) + λ∗ · Jg (x∗ ) = ∇Lx (x∗ ) = 0 2

In practice:

ˆ Do not forget the constraint qualification .

ˆ To not forget the constraint, write the FOC:

∂L ∗ ∗ ∂L ∗ ∗
(x , λ ) = 0 and (x , λ ) = 0 ∀i, j
∂xi ∂λj
as the batch of equations is the constraint.

ˆ When the constraint set can be defined implicitely globally (see proof), then
transform the problem into an unconstrained one.
APPENDIX A. MATHEMATICAL APPENDIX 48

To go further: Define the orthogonal space to Jg (x∗ ) (kernel of associated mapping):

Z(x∗ ) = {z ∈ Rn |Jg (x∗ ) · z = 0}

and the Hessian of L(·, λ∗ ) at x∗ :

m
X
∗ ∗ ∗
HL (x , λ ) = Hf (x ) + λ∗j Hgj (x∗ )
j=1

Consider x, x∗ feasible (i.e. g(x) = g(x∗ ) = 0) then by Taylor’s theorem:

L(x, λ∗ ) − L(x∗ , λ∗ ) = ∇x L(x∗ , λ∗ )(x − x∗ )
1
+ (x − x∗ )t HL,x (c, λ∗ )(x − x∗ )
2
which yields
1
f (x) − f (x∗ ) = (x − x∗ )t HL,x (c, λ∗ )(x − x∗ )
2
For SOSC, HL,x (c, λ ) should be negative definite on the set of vectors (x − x∗ ) such
∗

that g(x) = g(x∗ ) = 0, i.e. Z(x∗ ).

Theorem A.16 (Lagrange’s second order conditions) Let problem (A.2) with
f and g C 2 . Consider (x∗ , λ∗ ) a critical point of problem (A.2) such that x∗ ∈ IntX.
Then

1. if HL (x∗ , λ∗ ) is negative definite in Z(x∗ ), then x∗ is a strict local maximum;

2. if x∗ is a local maximum, then HL (x∗ , λ∗ ) is negative semidefinite in Z(x∗ ).

Interpretation of multiplicators: Consider for simplicity the one-constraint problem

(A.2) where g(x) = h(x) − b, and f and h are real-valued.
If they are C 2 , by the Implicit FTh: x∗ (b), λ∗ (b) are differentiable functions of b.
The FOC yield:
∂f ∗ ∂h ∗
(x (b)) = λ∗ (b) (x (b))
∂xi ∂xi
Let F (b) = f (x∗ (b)); by chain rule:
n n
′
X ∂f ∗ dxi ∗
X ∂h ∗ dxi
F (b) = (x (b)) (b) = λ (x (b)) = λ∗
i=1
∂x i db ∂x i db
|i=1 {z }
=1 since h(x∗ (b))=b

Hence λ∗ is the ‘shadow price’ of relaxing the constraint by one unit.

Consider an invertible n × n matrix A. The system of equations Ax = y has a

unique solution x = A−1 y. Can we have a similar result with a non-linear function
f ? Is there a solution to f (x) = y?
APPENDIX A. MATHEMATICAL APPENDIX 49

Proposition A.14 (Inverse function theorem) Let f : X ⊆ Rn → Rn be con-

tinuously differentiable in a neighborhood of x̄ ∈ IntX. Suppose det Jf (x̄) ̸= 0, then
there exists an open neighborhood U of x̄ and an open neighborhood V of f (x̄) such
that f (U ) = V and

ˆ the restriction of f to U , f |U is one-to-one, with inverse g : V → U ;

ˆ g is continuously differentiable on V and for any x ∈ U and y = f (x),

Dg(y) = Df (x)−1 [i.e. Jg (y) = Jf (x)−1 ]

Theorem A.17 (Implicit function theorem) Let f : X ⊆ Rn × Rm → Rm be

continuously differentiable in a neighborhood of (x̄, ȳ) ∈ IntX. Suppose

f (x̄, ȳ) = 0 and det Jf,y (x̄, ȳ) ̸= 0

   
∂fi
where Jf,y (x̄, ȳ) = (x̄, ȳ)
∂yj m×m
Then there exists an open neighborhood U of x̄ and a unique continuously differen-
tiable function g : U → Rm such that

ȳ = g(x̄) and ∀x ∈ U f (x, g(x)) = 0

Moreover
∀x ∈ U Jg (x) = −Jf,y (x, g(x))−1 Jf,x (x, g(x))

A.6 Convexity
Definition A.23 Let V be a real vector space; a subset C ⊆ V is convex if and
only if ∀ x, y ∈ C, t ∈ [0, 1] : tx + (1 − t)y ∈ C.

A convex combination: x = m
P Pm
i=1 λi xi where ∀i, λi ≥ 0 and i=1 λi = 1.

Lemma A.3 A convex set contains the convex comb. of any finite number of its
elements.

Proof: By induction. By def. C ⊃ convex comb. of any two elements. Suppose C ⊃

convex comb. of any m − 1 elements. Let x = m
P
i=1 λi xi , ∀i, xi ∈ C.

If λ1 = 1, x = x1 ∈ C - ended. If λ1 ̸= 1; x = λ1 x1 + (1 − λ1 ) m−1 i=1 1−λ1 xi . 2

λi
P

Example: Open and closed balls are convex.

Pn
The inner product (in finite dimension n): p · x = i=1 p i xi
APPENDIX A. MATHEMATICAL APPENDIX 50

Definition A.24 Let V be a finite dimensional real vector space, p ∈ V \{0} and
α ∈ R. Define:

ˆ hyperplane H(p, α) = {x ∈ V | p · x = 0}
(
H + (p, α) = {x ∈ V | p · x ≥ 0}
ˆ closed half-spaces
H − (p, α) = {x ∈ V | p · x ≤ 0}
(
intH + (p, α) = {x ∈ V | p · x > 0}
ˆ open half-spaces
intH − (p, α) = {x ∈ V | p · x < 0}

All these objects are convex; p is called a normal vector.

Proposition A.15 The intersection of arbitrarily many convex subsets of a real

vector space is convex.

Proof: Let (Ci )i∈I , I arbirary. Let C = ∩m

i=1 Ci ; let x, y ∈ C and t ∈ [0, 1]. Since
C ⊂ Ci and Ci convex, (1 − t)x + ty ∈ Ci , ∀i ⇒ ∈ C. 2

Example: Budget set B(p) = {x ∈ V | p · x ≥ w} = H − (p, α) ∩ (∩m +

i=1 H (ei , 0)) for
(ei ) canonical basis; it is convex.

A polyhedron in Rn is a bounded set that can be written as the intersection of

finitely many close half-spaces.

Definition A.25 Let V be a real vector space and A ⊆ V . The convex hull of A,
denoted coA, is the set of all convex combinations of finitely many elements of A.

Theorem A.18 (Carathéodory’s theorem) Let V be a real vector of dimension

n space and A ⊆ V . Any vector of coA can be written as a convex combination of
at most n + 1 vector of A.

Proof: Let A ̸= ∅ (otherwise: end). Let x ∈ coA; Prop ** 2/, ∃m ∈ N such that x
is the convex comb. of m elements of A: x = m
P
i=1 λi xi ; w.l.o.g. ∀i, λi > 0.
Let m be minimal. If m ≤ n + 1: end. If m > n + 1, the m − 1 vectors x2 − x1 ,
. . . , xm − x1 are necessarily linearly dependent: ∃α2 , . . . αm ∈ R, not all 0, such that
Pm Pm Pm Pm
i=2 αi (x i − x 1 ) = 0. Take α 1 = − i=2 α i : i=1 α i x i = 0, i=1 αi = 0.
λ
Obviously αj > 0 for some j. Let c = min{ αjj | αj > 0} and k such that
= c. Denote βj = λj − cαj : x = m
λk P Pm
αk j=1 βj xj and i=1 βi = 1 with ∀i, βi ≥ 0 and
βk = 0: hence x can be written as a convex comb. of less that m vectors of A - a
contradiction. 2
APPENDIX A. MATHEMATICAL APPENDIX 51

Definition A.26 Let V be a vector space. A finite subset A ⊆ V is affinely inde-

pendent in V if {z − x | z ∈ A\{x}} is linearly independent for any x ∈ A.

Lemma A.4 A is affinely indep. iff ∀α ∈ RA ,

P P
x∈A αx x = 0 and x∈A αx = 0 ⇒
α = 0.

Proof: (⇒) Suppose to the contrary that ∃α ∈ RA \{0},

P
x∈A αx x = 0 and
P
x∈A αx = 0: ∃x1 , x2 ∈ A, αx1 ̸= 0 and αx2 ̸= 0.
P P P
We have: 0 = x∈A αx x − ( x∈A αx )x1 = x∈A\{x1 } αx (x − x1 ). Therefore
x2 − x1 = x∈A\{x1 ,x2 } ααxx (x − x1 ) hence {x − x1 | x ∈ A\{x1 }} not lin. indep.
P
2
(⇐) Suppose A is not affinely indep.: ∃x1 , x2 ∈ A and β ∈ RA\{x1 ,x2 } \{0} s.t.
P P
x2 − x1 = x∈A\{x1 ,x2 } βx (x − x1 ); i.e. x∈A\{x1 } βx (x − x1 ) = 0 with βx2 = −1; i.e.
x∈A βx = 0 2
P P P
x∈A βx x = 0 with βx1 = − x∈A\{x1 } βx . Hence:

Lemma A.5 (Radon’s lemma) Let S ∈ Rn be nonempty and |S| = m ≥ n + 2,

there exists two disjoint sets A, B ⊆ S such that coA ∩ coB ̸= ∅.

Proof: Pick x1 , . . . , xm ∈ S. They are affinely dep. (there can be at most n + 1 affin.
indep. vectors in Rn ): ∃α ∈ Rm \{0},
Pm Pm
i=1 α i x i = 0 and i=1 αi = 0.
Let I = {i : αi > 0}, J = {i : αi < 0}, A = {xi : i ∈ I} and B = {xi : i ∈ J}.
Define for i ∈ I: βi = Pαiαi , then I βi xi ∈ coA.
P
I

Moreover I βi xi = P 1 αi I αi xi = − P1 αi (− J αi xi ) ∈ coB. 2
P P P
I J

Theorem A.19 (Helly’s theorem) If C is a finite class of convex subsets of Rn

such that |C| ≥ n + 1 and ∩S =
̸ ∅ for any S ⊆ C with |S| = n + 1, then ∩C ̸= ∅.

Proof: By induction on |C|. If |C| = n + 1: ok. Suppose ok if |C| = m ≥ n + 1. Take

|C| = m + 1; by induction hypothesis: ∃xi ∈ ∩(C\{Ci }), i = 1, . . . , m + 1. If xi = xj
for some i ̸= j: done. Otherwise, let T = {x1 , . . . , xm+1 }; |T | ≥ n + 2; by Radon’s
lemma: ∃A, B ⊆ T disjoint such that coA ∩ coB ̸= ∅.
Take x ∈ coA ∩ coB; we show x ∈ ∩C. Pick i; we show: x ∈ Ci . If xi ∈ A,
then xi ∈
/ B ⇒ Ci ⊇ B (the only element of T not in Ci is xi ) ⇒ Ci ⊇ coB ∋ x.
/ A ⇒ Ci ⊇ A ⇒ Ci ⊇ coA ∋ x. 2
Similarly, if xi ∈ B, then xi ∈

For infinite class, we need compactness. Th **: any nonempty finite subset of the
class C has nonempty intersection. By Prop **: ∩C ̸= ∅.

Proposition A.16 Let (V, ∥ · ∥) be a normed vector space and A ⊆ V be open, then
coA is open.
APPENDIX A. MATHEMATICAL APPENDIX 52

Proof: Since A ⊆ coA and A open, A ⊆ int coA. But int coA is convex ⇒ coA ⊆
int coA, hence coA = int coA. 2

Remark: The convex hull of a closed set is not always closed. Let A = {(x, 0) | x ∈
R} ∪ {(0, 1)} ⊆ R2 . A is closed, but coA = {(x, y) | x ∈ R, 0 ≤ x < 1} ∪ {(0, 1)} ⊆ R2
is not.

Theorem A.20 (Mazur’s theorem) Let K ⊆ Rn be compact, then coK is com-

pact.

Proof: Define f : Rn+1 × (Rn )n+1 → Rn by f (λ, x1 , . . . , xn+1 ) =

Pn+1
i=1 λi xi .
Let ∆n+1 be the n-dimensional simplex (which is compact - Heine-Borel) and
A = ∆n+1 × K n+1 . By Tychonov’s theorem, A is compact.
By Carathéodory’s theorem, coK = f (A). Since f is continuous and A compact,
coK is compact. 2

Proposition A.17 Let A ⊆ Rn be such that 0 ∈ intA. Then the mapping mA :

Rn → R given by mA (x) = inf{α > 0 | x/α ∈ A} is well-defined; it is called the
Minkowski functional of A.

Proof: Since 0 ∈ intA, ∃ϵ > 0 : B(0, ϵ) ⊆ A. Hence ∀x ∈ Rn , x/α ∈ A if (e.g.)

α = 2∥x∥/ϵ. Hence {α > 0 | x/α ∈ A} is nonempty and bounded from below (by
0), therefore the infimum exists (by the Supremum Axiom). 2

Definition A.27 Let H = H(p, α) be a hyperplane in Rn , p ∈ Rn \{0} and α ∈ R,

A, B ⊆ Rn . We say that:

ˆ H separates A and B if A ⊆ H + (p, α) and B ⊆ H − (p, α);

ˆ H strongly separates A and B if ∃ϵ > 0, A ⊆ H + (p, α + ϵ) and B ⊆

H − (p, α − ϵ);

Exercise: Let xk = (k, 0) ∈ R2 , k = 0, 1, 2, 3. Let A = [x0 , x1 ] and B = [x2 , x3 ].

1. Show that H((0, 1), 0) separates A and B.

2. Find a hyperplane which strongly separates A and B.

Theorem A.21 (Separating hyperplane theorem) Let A, B ⊆ Rn be nonempty

convex sets.
APPENDIX A. MATHEMATICAL APPENDIX 53

ˆ If A ∩ B = ∅, then A and B can be separated.

ˆ If A ∩ B = ∅, A compact and B closed, they can be strongly separated.

Let H(p, α) with p ∈ Rn \{0} and α ∈ R. For any x∗ ∈ H(p, α), H(p, α) is the
hyperplane passing through x∗ :

H(p, x∗ ) = {x ∈ Rn | p · (x − x∗ ) = 0}.

Theorem A.22 (Supporting hyperplane theorem) Let C ⊆ Rn be a convex

set with nonempty interior, and let x∗ ∈ bdC. Then ∃p ∈ Rn \{0} such that C ⊆
H − (p, x∗ ).

Proof: Since intC nonempty and convex + {x∗ } convex and disjoint from intC, by
Cor ** ∃p ∈ Rn \{0} : p · c ≤ p · x∗ ∀c ∈ cl intC = clC. QED. 2

Example: For utility function U , upper contour set U≺ (x) = {y ∈ X | x ≺ y} is

convex; bdU≺ (x) is the indifference surface through x; in part. x ∈ bdU≺ (x). There
exists a supporting plane with normal vector p interpreted as a price vector. Hence
any point strictly preferred to x is more expensive than x: x is optimal.

Lemma A.6 (Farkas’ lemma) Let A ∈ Rmn and c ∈ Rn . Then one and only one
of the following systems has a solution
 
Ax ≤ 0m  
 y ≥ 0m 


t
cx>0 (I) At y = c (II)
x ∈ Rn  y ∈ Rm

 



Proof: Suppose (II) has a solution y ∗ . For any x ∈ Rn such that Ax ≤ 0m , we have
y t Ax = (At y)t x = ct x. Since y ≥ 0m and Ax ≤ 0m , we get ct x ≤ 0 hence (I) cannot
have a solution.
Suppose now that (II) does not have a solution: c is not in the set

S = {z ∈ Rn | z = At y for some y ∈ Rm , y ≥ 0m }

which is closed, convex, nonempty (it is the cone generated by A’s row vectors).
Cor **: S and {c} can be strongly separated: ∃p ∈ Rn \{0n } and α ∈ R such
that ∀z ∈ S, p · z < α < p · c. Since 0n ∈ S, 0 < α and α < p · c = ct p >.
APPENDIX A. MATHEMATICAL APPENDIX 54

We claim: Ap ≤ 0m so that p is a solution of (I). Suppose not: ∃i: (Ap)i > 0. Let
2α
y = (0, . . . , 0, yi , 0, . . . , 0) with yi = (Ap)i
. At y ∈ S and α > p · At y = (Ap)t y = 2α
− a contradiction to α > 0. 2

Geometry for n = m = 2: (I) has a solution if the intersection of the half-spaces:

a1 x ≤ 0, a2 x ≤ 0 and ct x > 0 is nonempty (where a1 , a2 are the row vectors of A).
This is equivalent to say that c does not lie in the cone generated by a1 , a2 .

Theorem A.23 (Brouwer’s fixed point theorem) Let C ⊆ Rn be nonempty,

compact and convex. Then any continuous function f : C → C has a fixed point.