2.

Separating Hyperplane Theorems

Daisuke Oyama

Mathematical Economics

This version: June 19, 2023

Things to know from real analysis
▶ For S ⊂ R,
▶ α = max S if
▶ α ≥ x for all S and
▶ α ∈ S.
▶ α = min S if
▶ α ≤ x for all S and
▶ α ∈ S.
▶ α = sup S if
▶ α ≥ x for all S and
▶ if β ≥ x for all S, then β ≥ α.
▶ α = inf S if
▶ α ≤ x for all S and
▶ if β ≤ x for all S, then β ≤ α.
▶ sup S exists if S is bounded above;
inf S exists if S is bounded below.

1 / 61
▶ Euclidean norm: for x = (x1 , . . . , xn ) ∈ Rn ,
√∑n
kxk = 2
i=1 (xi ) .

▶ Euclidean distance: for x, y ∈ Rn , d(x, y) = kx − yk.

▶ A sequence {xk } in Rn converges to x0 ∈ Rn if for any ε > 0,
there exists a natural number K such that d(xk , x0 ) < ε for
all k ≥ K.
In this case,
▶ {xk } is said to be convergent,
▶ x0 is called the limit of {xk }, and
▶ we write limk→∞ xk = x0 or xk → x0 as k → ∞.

▶ A sequence {xk } in Rn is called a Cauchy sequence if

for any ε > 0, there exists a natural number K such that
d(xk , xℓ ) < ε for all k, ℓ ≥ K.
▶ A sequence {xk } in Rn is convergent if and only if
it is a Cauchy sequence.
2 / 61
▶ S ⊂ Rn is closed if for any convergent sequence in {xk } in S
with xk → x∗ , we have x∗ ∈ S.
▶ S ⊂ Rn is open if for every x ∈ S, there exists ε > 0 such
that Bε (x) ⊂ S,
where Bε (x) = {y ∈ Rn | d(y, x) < ε}.
▶ S ⊂ Rn is closed if and only if Rn \ S is open.
▶ Basic properties:
▶ ∅ and Rn are both closed and open.
▶ The union of any family of open sets is open.
▶ The intersection of a finite number of open sets is open.
▶ The intersection of any family of closed sets is closed.
▶ The union of a finite number of closed sets is closed.

3 / 61
▶ x ∈ S is an interior point of S if there exists ε > 0 such that
Bε (x) ⊂ S.
The set of all interior points of S is called the interior of S
and denoted int S.
▶ int S is the largest open set that is contained in S.

▶ x ∈ Rn is a boundary point of S if for any ε > 0,

Bε (x) ∩ S 6= ∅ and Bε (x) ∩ (Rn \ S) 6= ∅.
The set of all boundary points of S is called the boundary of
S and denoted bd S.
▶ The closure of S ⊂ Rn is the set of all points that are
the limits of convergent sequences of points in S and denoted
cl S.
▶ cl S is the smallest closed set that contains S.

4 / 61
▶ Relationships:
▶ cl S = Rn \ int(Rn \ S)
▶ int S = Rn \ cl(Rn \ S)
▶ bd S = cl S \ int S

5 / 61
▶ S ⊂ Rn is bounded if there exists r > 0 such that S ⊂ Br (0).

▶ S ⊂ Rn is compact if it is closed and bounded.

▶ If S ⊂ R is compact, then sup S ∈ S and inf S ∈ S

(and hence sup S = max S and inf S = min S).
▶ If S ⊂ Rn is bounded, then any sequence in S has
a convergent subsequence.
If in addition S is closed (i.e., S is compact),
then any sequence in S has a convergent subsequence and its
limit is in S.
▶ Conversely, if every sequence in S has a convergent
subsequence whose limit is in S, then S is compact.

6 / 61
▶ A family of subsets of Rn is said to have the finite intersection
property if the intersection of any finite subfamily of it is
nonempty.
▶ For S ⊂ Rn , the following conditions are equivalent:
▶ S is compact.
▶ For any family (Fλ )λ∈Λ of closed subsets of S that has
∩
the finite intersection property, we have λ∈Λ Fλ 6= ∅.

7 / 61
▶ Let S ⊂ Rn , S 6= ∅.
A function f : S → Rm is continuous at x̄ ∈ S if for any
ε > 0, there exists δ > 0 such that

d(x, x̄) < δ, x ∈ S ⇒ d(f (x), f (x̄)) < ε.

▶ f : S → Rm is continuous on T ⊂ S if it is continuous at
every x ∈ T .
▶ f : S → Rm is continuous if it is continuous on S.

▶ f : S → Rm is continuous at x̄ ∈ S if and only if

for any sequence {xk } in S such that xk → x̄, we have
f (xk ) → f (x̄).
▶ Examples:
▶ (x, y) 7→ d(x, y) is continuous.
▶ For A ∈ Rm×n , x 7→ Ax is continuous.

8 / 61
Proposition 2.1 (Weierstrass’ Theorem)
Suppose that S ⊂ Rn , S 6= ∅, is compact and f : S → R is
continuous. Then maxx∈S f (x) and minx∈S f (x) exist,
i.e., there exist x∗ , x∗∗ ∈ S such that f (x∗∗ ) ≤ f (x) ≤ f (x∗ ) for
all x ∈ S.

9 / 61
Convex Sets

Definition 2.1
C ⊂ Rn is convex if for any x, y ∈ C and λ ∈ [0, 1], we have
λx + (1 − λ)y ∈ C.

10 / 61
Properties of Convex Sets

Proposition 2.2
Suppose that C, D ⊂ Rn are convex.
▶ C + D = {x + y | x ∈ C, y ∈ D} is convex.

▶ For α ∈ R, αC = {αx | x ∈ C} is convex.

Proposition 2.3
The intersection of any family of convex sets is convex.

11 / 61
Proposition 2.4
If C ⊂ Rn is convex, then cl C is also convex.

Proof
▶ Bε (0) = {y ∈ Rn | kyk < ε} is convex.
∩
▶ Then cl C = ε>0 (C + Bε (0)) is convex if C is convex.

Proposition 2.5
If C ⊂ Rn is convex, then int C is also convex.

12 / 61
▶ For S ⊂ Rn , the set of all convex combinations of finite
subsets of S is called the convex hull of S and denoted by
conv(S).
▶ conv(S) is the smallest convex set that contains S.

13 / 61
Carathéodory’s Theorem

Proposition 2.6 (Carathéodory’s Theorem)

1. For S ⊂ Rn , S 6= {0}, each x ∈ cone(S) is written as a conic
combination of linear independent elements of S.

2. For S ⊂ Rn , each x ∈ conv(S) is written as a convex

combination of at most n + 1 elements of S.

Proof of Part 1
▶ Immediate from Proposition 1.6 (Fundamental Theorem of
Linear Inequalities).

14 / 61
Proof of Part 2

▶ Let x ∈ conv(S).
∑
Then we have x = Jj=1 λj xj for some x1 , . . . , xJ ∈ S and
∑
λ1 , . . . , λJ ≥ 0, Jj=1 λj = 1.

▶ Consider T = {(x1 , 1), . . . , (xJ , 1)} ⊂ Rn+1 .

Then (x, 1) ∈ cone(T ).
▶ By part 1, there is an LI subset T ′ ⊂ T such that
∑
(x, 1) = j∈T ′ µj (xj , 1) with µj ≥ 0, where |T ′ | ≤ n + 1.
▶ From the 1st through nth coordinates we have
∑
x = j∈T ′ µj xj , while from the (n + 1)st coordinate we have
∑
j∈T ′ µj = 1.

15 / 61
Convex Hull of a Compact Set

Proposition 2.7
If S ⊂ Rn is bounded, then cl(conv(S)) = conv(cl(S)).
In particular, if S is compact, then conv(S) is compact.

16 / 61
Proof
▶ Since conv(S) ⊃ S, we have cl(conv(S)) ⊃ cl(S).
Since cl(conv(S)) is convex (Proposition 2.4), we have
cl(conv(S)) ⊃ conv(cl(S)).
▶ Since S ⊂ cl(S), we have conv(S) ⊂ conv(cl(S)).
We want to show that conv(cl(S)) is closed if S is bounded.
▶ Let {xk } ⊂ conv(cl(S)), and assume xk → x̄.

▶ By Carathéodory’s Theorem (Proposition 2.6 part 2),

each xk is written as

xk = α1k xk,1 + · · · + αn+1

k
xk,n+1 ,

where
∑
▶ (α1k , . . . , αn+1
k
) ∈ ∆ = {α ∈ Rn+1 | αi ≥ 0, i αi = 1},
▶ x k,1
,...,x k,n+1
∈ cl(S).
17 / 61
▶ Since ∆ and cl(S) are compact, there exists a sequence
k(ℓ)
{k(ℓ)} such that the limits ᾱi = limℓ→∞ αi and
x̄i = limℓ→∞ xk(ℓ),i exist where (ᾱ1 , . . . , ᾱn+1 ) ∈ ∆ and
x̄1 , . . . , x̄n+1 ∈ cl(S).
▶ Hence,

x̄ = ᾱ1 x̄1 + · · · + ᾱn+1 x̄n+1 ,

so that x̄ ∈ conv(cl(S)).

18 / 61
Separating Hyperplane Theorems

▶ The textbook proves the strict separating hyperplane theorem

from scratch.
▶ It then states the weak separating hyperplane theorem without
proof, saying “The proof is similar to the previous one.”
(In fact, the proof is far from “similar”.)
▶ Here, we prove the weak separating hyperplane theorem by
Farkas’ Lemma (which we proved by an algebraic argument).
▶ Then we prove the strict version from the weak version.

19 / 61
Weak Separating Hyperplane Theorem

Proposition 2.8 (Weak Separating Hyperplane Theorem)

Suppose that C ⊂ Rn is a convex set, and that b ∈
/ C.
Then there exists h ∈ R , h 6= 0 such that
n

hx ≤ hb for all x ∈ C.

▶ The proof below is an adoption of a proof in some lecture notes by

Atsushi Kajii (which proves this theorem from the strict version).
▶ A similar argument (similar to Kajii’s) is also found in Berkovitz,
Convexity and Optimization in Rn , Chapter II, Theorem 3.2.

20 / 61
Proof
▶ Write P 0 = {h ∈ Rn | khk = 1}, which is compact.

▶ Let C ⊂ Rn be convex and b ∈

/ C.
▶ For each x ∈ C, let

Px = {h ∈ P 0 | hx ≤ hb},

which is a closed subset of P 0 .

∩
We want to show that x∈C Px 6= ∅.
▶ We show that the family {Px }x∈C of closed subsets of
compact set P 0 has the finite intersection property.
Take any x1 , . . . , xm ∈ C.
[ ]
Write A = x1 · · · xm ∈ Rn×m .

21 / 61
▶ Since b ∈
/ conv(A) (⊂ C), there exists no α ∈ Rm such that
b = Aα, 1α = 1, and [ ]α ≥ [0 (where
] 1 ∈ Rm is the vector of
b A
ones), or such that = T α and α ≥ 0.
1 1
▶ Then by Farkas’
[ ]Lemma, there exist
[ h ] ∈ R and k ∈ R such
n
[ ] A [ ] b
that h k ≤ 0 and h k > 0, or
1T 1
hxj ≤ −k < hb for all j = 1, . . . , m,
∩
so that h ∈ mj=1 Pxj .
∩m
▶ Thus, j=1 Pxj 6= ∅.
∩
▶ Hence, by the compactness of P 0 , we have x∈C Px 6= ∅, as
desired.

22 / 61
Proposition 2.9
Suppose C, D ⊂ Rn , C, D 6= ∅, are convex, and that C ∩ D = ∅.
Then there exists h ∈ Rn , h 6= 0 such that

hx ≤ hy for all x ∈ C and y ∈ D.

23 / 61
Proof

▶ Let K = C − D (= {x − y | x ∈ C, y ∈ D}). Then

▶ K 6= ∅ (∵ C, D 6= ∅);
▶ K is convex (∵ C and D convex);
▶ 0∈
/ K (∵ C ∩ D = ∅).

▶ Therefore, by the weak separating hyperplane theorem, there

exists h ∈ Rn , h 6= 0, such that

hz ≤ h0 for all z ∈ K,

or

hx ≤ hy for all x ∈ C and y ∈ D.

24 / 61
Strict Separating Hyperplane Theorem

Proposition 2.10 (Strict Separating Hyperplane Theorem)

Suppose that C ⊂ Rn is a closed convex set, and that b ∈
/ C.
Then there exist h ∈ Rn , h 6= 0, and β ∈ R such that

hx ≤ β < hb for all x ∈ C.

25 / 61
Proof

▶ Let b ∈
/ C.
▶ By the closedness of C, there exists ε̄ > 0 such that
C ∩ Bε̄ (b) = ∅.
▶ By the convexity of C (and Bε̄ (b)), it follows from
Proposition 2.9 that there exists h ∈ Rn , h 6= 0 such that
hx ≤ hy for all C and all y ∈ Bε̄ (b).
Normalize h so that khk = 1.
▶ Letting y = b − 2ε̄ h, we have hx ≤ hb − ε̄
2 for all x ∈ C,
where hb − 2ε̄ < hb.
▶ Finally, let β = hb − 2ε̄ .

26 / 61
Proposition 2.11
Suppose C, D ⊂ Rn , C, D 6= ∅, are convex and closed, and that
C ∩ D = ∅.
If C or D is bounded, then there exist h ∈ Rn , h 6= 0, and β ∈ R
such that

hx < β < hy for all x ∈ C and y ∈ D.

▶ The boundedness of C or D is indispensable.

27 / 61
Proof

▶ Let K = C − D (= {x − y | x ∈ C, y ∈ D}). Then

▶ K 6= ∅ (∵ C, D 6= ∅);
▶ K is convex (∵ C and D convex);
▶ 0∈
/ K (∵ C ∩ D = ∅).

▶ Suppose that C is bounded and hence is compact by

the closedness of C.
We want to show that K is closed.

28 / 61
▶ Take any sequence {z k } in K, and assume that z k → z ∗ .
▶ For each k, let xk ∈ C and y k ∈ D be such that z k = xk − y k .
▶ By the compactness of C, there are a subsequence of {xk }
(again denoted {xk }) and x∗ ∈ C such that xk → x∗ .
▶ Then y k = xk − z k converges to some y ∗ , where y ∗ ∈ D by
the closedness of D.
▶ Then we have z k = xk − y k → z ∗ = x∗ − y ∗ , and hence
z ∗ ∈ K.
▶ This proves that K is closed.

29 / 61
▶ Therefore, by the strict separating hyperplane theorem, there
exist h ∈ Rn , h 6= 0, and β ′ ∈ R such that

hz < β ′ < h0 for all z ∈ K,

or

hx < hy + β ′ < hy for all x ∈ C and y ∈ D.

▶ Then let, for example, β = inf y∈D hy + β′

2.

30 / 61
Extreme Points and Extreme Rays

Definition 2.2
For S ⊂ Rn , x ∈ S is an extreme point of S if

x = λy + (1 − λ)z, y, z ∈ S, λ ∈ (0, 1) =⇒ y = z = x.

Definition 2.3
For S ⊂ Rn ,
r ∈ S, r 6= 0, is a ray of S if x + λr ∈ S for all x ∈ S and λ ≥ 0;
r ∈ S is an extreme ray of S if

r = λu + (1 − λ)v, u, v: rays of S, λ ∈ (0, 1)

=⇒ u = αv for some α > 0.

31 / 61
Krein-Milman Theorem

▶ Denote the set of extreme points of C by ext(C).

Proposition 2.12 (Krein-Milman Theorem)

Let C ⊂ Rn , C 6= ∅, be a compact convex set.
Then C = conv(ext(C)).

▶ In the proof given in the textbook, I could not prove the closedness
of K from the induction hypothesis (rather than proving
the Krein-Milman Theorem itself by a different proof).

32 / 61
 ▶ We prove the theorem in a stronger form.

Proposition 2.13 (Krein-Milman Theorem)

Let C ⊂ Rn , C 6= ∅, be a compact convex set.
Then each x ∈ C is written as a convex combination of at most
n + 1 extreme points of C.

▶ The proof is by induction on the dimension of C.

33 / 61
Dimension of a set
Definition 2.4
{x0 , . . . , xm } ⊂ Rn is affinely independent if
{x1 − x0 , . . . , xm − x0 } is LI.

Definition 2.5
For S ⊂ Rn , the dimension of S, dim S, is the largest number m
for which S contains some affinely independent vectors x0 , . . . , xm .

▶ For any x0 ∈ S, dim(S) = rank(S − {x0 }).

▶ dim Rn = n (take 0 and the unit vectors e1 , . . . , en ).

▶ For a hyperplane H ⊂ Rn , i.e., H = {x ∈ Rn | hx = β} for
some h ∈ Rn , h 6= 0, and β ∈ R, dim H = n − 1.
▶ For any x0 ∈ H, let H 0 = H − {x0 } = {x ∈ Rn | hx = 0}.
▶ H 0 = ker(hT ) and rank(span(h)) = 1, and hence
dim(H) = rank(H 0 ) = n − 1.

34 / 61
Proof of Proposition 2.13

▶ We prove by induction on the dimension of C.

▶ If dim(C) = 0, where C is a singleton set, the statement is

obviously true.
▶ Assume that the statement is true for any compact convex set
C with dim(C) ≤ m − 1.
▶ Suppose that dim(C) = m. Denote K = conv(ext(C)).
We can embed C into Rm , so that we can assume C ⊂ Rm
(where the structure of convex combinations does not
change).

35 / 61
Claim 1
Each x ∈ bd C is written as a convex combination of at most m
extreme points of C.

Proof
▶ Take any x̄ ∈ bd C.
By the weak separating hyperplane theorem (applied to int C
which is convex), there exists a hyperplane
H = {x ∈ Rm | hx = hx̄} such that hy ≤ hx̄ for all y ∈ C.
▶ Since C ∩ H is compact and convex and
dim(C ∩ H) ≤ m − 1, by the induction hypothesis there are
m extreme points y 1 , . . . , y m of C ∩ H such that
x̄ ∈ conv({y 1 , . . . , y m }).
▶ We want to show that y 1 , . . . , y m are extreme points of C.

36 / 61
▶ Let y i = λz + (1 − λ)w, z, w ∈ C, and λ ∈ (0, 1).

▶ Then we have

hx̄ = hy i = λhz + (1 − λ)hw

≤ λhx̄ + (1 − λ)hx̄ = hx̄.

Thus, the inequality in fact holds with equality, and hence,

hz = hw = hx̄, meaning that z, w ∈ H.
▶ Since y i is an extreme point of C ∩ H, it must be that
z = w = yi.
This proves that each y i is an extreme point of C.
[End of the proof of Claim 1]

37 / 61
▶ Claim 1 in particular implies that ext(C) 6= ∅.
Fix any x0 ∈ ext(C).
▶ Take any x ∈ C.
If x = x0 , we are done, so assume that x 6= x0 .
▶ Let α0 = max{α ∈ R | x0 + α(x − x0 ) ∈ C} ≥ 1, which is
well defined by the compactness of C.
Then y = x0 + α0 (x − x0 ) ∈ bd C.
▶ Then by Claim 1, there exist x1 , . . . , xm ∈ ext(C) such that
∑ ∑m
y= m i=1 αi x for some α1 , . . . , αm ≥ 0 with
i
i=1 αi = 1.
▶ Then we have
1
x = x0 + (y − x0 )
α0
α0 − 1 0 1 α0 − 1 0 1 ∑
= x + 0y = x + 0 m i
i=1 αi x ,
α0 α α0 α
∑
where αα0 −1
0
, α10 αi ≥ 0 and αα0 −1
0
+ α10 m
i=1 αi = 1, as desired.
38 / 61
Application: Walrasian Equilibrium in Exchange Economies
▶ Goods 1, . . . , n
▶ Agents A = {1, . . . , m}
▶ For each agent i ∈ A:
▶ Endowment wi ∈ Rn+
Assume wi  0.
▶ Utility function U i : Rn+ → R
Assumed to be
▶ continuous;
▶ strictly quasi-concave; and
▶ strictly increasing: i.e., if y ≥ x and y ̸= x, then
U i (y) > U i (x).
∑
▶ Let M ∈ Rn+ be such that M ≥ i∈A wi .
(In particular, M  wi for all i ∈ A.)
▶ p ∈ Rn+ : Price vector (to be determined in equilibrium)
39 / 61
▶ Demand function of agent i:

di (p) = arg max{U i (x) | x ∈ Rn+ , px ≤ pwi , x ≤ M }

▶ “x ≤ M ” is a non-standard constraint, which makes the

domain compact even when the prices of some goods are zero.
▶ By the continuity of U i , the right-hand side is nonempty.
▶ By the strict quasi-concavity of U i , the right-hand side is
a singleton set.
→ We regard di (p) as a function (instead of correspondence).

40 / 61
Observation 1
For any p ∈ Rn+ , if px < pwi and x ≤ M , then U i (x) < U i (di (p));
in particular, pdi (p) = pwi .

▶ This holds if U i satisfies local insatiability within

{x ∈ Rn+ | x ≤ M } (denote this set by M ),
i.e., the property that for any x ∈ M and any ε > 0,
there exists x′ ∈ M such that kx′ − xk < ε and
U i (x′ ) > U i (x).
(Local insatiability within Rn+ is not sufficient.)

41 / 61
Proof

▶ Let p 6= 0.

▶ Let x ∈ Rn+ be such that px < pwi and x ≤ M .

We want to show that such an x is not optimal.
▶ By M  wi , there must exist j such that pj > 0 and
xj < Mj (otherwise we would have px > pwi ).
▶ Let x′ ∈ Rn+ be such that x′j is slightly larger than xj (while
x′k = xk for all k 6= j) so that we still have px′ ≤ pwi and
x′j ≤ Mj .

▶ By strict monotonicity of U i , we have U i (x′ ) > U i (x).

This means that x is not optimal.

42 / 61
Observation 2
di (tp) = di (p) for any t > 0.

Observation 3
If U i (x) > U i (di (p)) and x ≤ M , then px > pwi .

∑
▶ We normalize a price vector p ≥ 0, p 6= 0, so that j pj = 1,
∑
or consider p as an element of ∆ = {p ∈ Rn+ | j pj = 1}.

43 / 61
Definition 2.6
A pair of price vector p ∈ Rn+ and allocation
X = (x1 , . . . , xm ) ∈ (Rn+ )m is a Walrasian equilibrium if
▶ [utility maximization]
xi = di (p) for all i ∈ A, and
▶ [market clearing]
∑ i
∑ i
i∈A x = i∈A w .

▶ The market clearing condition should be imposed as

∑ ∑
an inequality (i.e., i∈A xi ≤ i∈A wi ) if we do not assume
monotonicity of U i .

44 / 61
Lemma 2.14
di (p) is continuous on ∆.

▶ By the continuity of U i and the “continuity” of the constraint

correspondence p 7→ {x ∈ Rn+ | px ≤ pwi , x ≤ M }.

45 / 61
Proof

▶ Let {pk } be a sequence in ∆ and assume that pk → p∗ ∈ ∆.

▶ Write xk = di (pk ).
Since it is contained in the compact set {x ∈ Rn+ | x ≤ M },
we assume that {xk } is convergent with limit x∗ ∈ Rn+ ,
x∗ ≤ M .
We want to show that di (p∗ ) = x∗ .
▶ First, since pk xk ≤ pk wi for all k,
by k → ∞ we have p∗ x∗ ≤ p∗ wi .

46 / 61
▶ Second take any z ∈ Rn+ such that p∗ z ≤ p∗ wi and z ≤ M .
We want to show that U i (z) ≤ U i (x∗ ).
▶ For any ε > 0, let z ε ∈ Rn+ be such that kz ε − zk < ε,
p∗ z ε < p∗ wi , and z ε ≤ M .
(Note that p∗ wi > 0 since wi  0 by assumption.)
▶ Let K be such that pk z ε ≤ pk wi for all k ≥ K.

▶ Then by optimality we have U i (z ε ) ≤ U i (xk ).

▶ Letting k → ∞, we have U i (z ε ) ≤ U i (x∗ ) by continuity of U i .

▶ Finally, letting ε → 0, we have U i (z) ≤ U i (x∗ ) again by

continuity of U i .

47 / 61
 ∑ ∑
▶ Define the function E(p) = i
i∈A d (p) − i∈A w
i.

· · · Excess demand function

Continuous by Lemma 2.14.

Lemma 2.15 (Walras’ Law)

For any p ∈ Rn+ , pE(p) = 0.

▶ By Observation 1.

48 / 61
Existence of Walrasian equilibrium

Proposition 2.16
There exists a Walrasian equilibrium.

▶ p ∈ ∆ is a Walrasian equilibrium price vector if and only if

E(p) = 0, or it is a fixed point of the function p + E(p).
▶ But p + E(p) ∈
/ ∆ in general.
We will modify this function so that the value is in ∆.
▶ Then use Brouwer’s Fixed Point Theorem.

49 / 61
Brouwer’s Fixed Point Theorem

Proposition 2.17
Suppose that X ⊂ RN is a nonempty, compact, and convex set,
and that f : X → X is a continuous function from X into itself.
Then f has a fixed point, i.e., there exists x ∈ X such that
x = f (x).

50 / 61
Proof of Proposition 2.16

▶ Write Ej+ (p) = max{Ej (p), 0}, which is continuous in p.

▶ Define the function f : ∆ → ∆ by

pj + Ej+ (p)
fj (p) = ∑ ,
1+ m +
j=1 Ej (p)

which is continuous, mapping the compact set ∆ to itself.

▶ By Brouwer’s Fixed Point Theorem, f has a fixed point p ∈ ∆:

pj + Ej+ (p)
pj = ∑ .
1+ m +
j=1 Ej (p)

51 / 61
▶ Then by Walras’ Law pE(p) = 0, we have
∑ ∑ +
∑ j pj Ej (p) + E (p)Ej (p)
0= pj Ej (p) = ∑ j+ j
1 + j E j(p)
j
∑ +
j Ej (p)Ej (p)
= ∑ ,
1 + j E + j(p)
∑
and therefore j Ej+ (p)Ej (p) = 0.
▶ Since
{
Ej (p)2 if Ej (p) > 0,
Ej+ (p)Ej (p) =
0 if Ej (p) ≤ 0,

it must be that Ej (p) ≤ 0 for all j.

52 / 61
▶ Finally, we want to show that Ej (p) = 0 for all j (by strict
monotonicity of U i ).
∑
▶ By Walras’ Law, j pj Ej (p) = 0, where pj Ej (p) ≤ 0 as
shown.
▶ If Ej (p) < 0, then pj = 0,
but by monotonicity
∑ of U i , we would have
dij (p) = Mj ≥ h∈A wjh (> 0) for all i ∈ A, which violates
Ej (p) ≤ 0.

53 / 61
Pareto Efficiency of Walrasian equilibrium

▶ An allocation X = (x1 , . . . , xm ) ∈ (Rn+ )m is feasible if

∑ ∑
i∈A x ≤
i i
i∈A w .
▶ An allocation Y Pareto dominates an allocation X if
▶ U i (y i ) ≥ U i (xi ) for all i ∈ A, and
▶ U i (y i ) > U i (xi ) for some i ∈ A.

▶ A feasible allocation X is Pareto efficient (or Pareto optimal)

if there exists no feasible allocation Y that Pareto dominates
X.

54 / 61
First Fundamental Theorem of Welfare Economics

Proposition 2.18
If (p, X) is a Walrasian equilibrium, then X is Pareto efficient.

▶ Uses only Observation 1.

55 / 61
Proof
▶ Suppose that an allocation Y Pareto dominates X, i.e.,

U i (y i ) ≥ U i (xi ) for all i ∈ A, (1)

U i (y i ) > U i (xi ) for some i ∈ A. (2)

We want to show that Y is not feasible.

▶ If y i 6≤ M for some i ∈ A, then clearly Y is not feasible.
Suppose that y i ≤ M for all i ∈ A.
▶ By (1) and Observation 1, we have

py i ≥ pwi for all i ∈ A.

▶ By (2), we have

py i > pwi for some i ∈ A.

56 / 61
▶ Therefore, we have
( )
∑ ∑ ∑
p yi − wi = (py i − pwi ) > 0.
i∈A i∈A i∈A

∑ ∑
▶ This implies that i∈A y i ≤ i∈A wi does not hold,
i.e., Y is not feasible,
(∑ ∑ )
i∈A y − i∈A w ≤ 0 otherwise.
for, we would have p i i

57 / 61
Second Fundamental Theorem of Welfare Economics

Proposition 2.19
Suppose that X = (w1 , . . . , wm ) is Pareto efficient. Then there
exists p ∈ Rn+ such that (p, X) is a Walrasian equilibrium.

▶ Uses
▶ quasi-concavity,
▶ local insatiability, and
▶ continuity of U i ; and
▶ wi  0.

58 / 61
Proof
▶ Define

Ŝ i = {y i ∈ Rn+ | U i (y i ) > U i (wi )},

∑
and define Ŝ = i∈A Ŝ i , which is a convex set by
the quasi-concavity of U i ’s.
▶ By the Pareto efficiency of X = (w1 , . . . , wm ),
∑
Ŝ ∩ ({ i∈A wi } − Rn+ ) = ∅.
▶ By the weak separating hyperplane theorem, there exists
p ∈ Rn , p 6= 0, such that
(∑ )
py ≥ p i∈A w − z for all y ∈ Ŝ and z ≥ 0.
i

▶ Since this holds for all z ≥ 0, it must be that p ≥ 0.

▶ We want to show that (p, X) is a Walrasian equilibrium.

59 / 61
▶ Fix any i ∈ A.
Suppose that y i ∈ Rn+ , U i (y i ) > U i (wi ), and y i ≤ M .
▶ For each j 6= i, by strict monotonicity of U j (local insatiability
is sufficient) we have y j arbitrarily close to wj such that
U j (y j ) > U i (wi ).
∑
▶ Then j y j ∈ Ŝ, and therefore,
∑ ∑
p(y i + j̸=i y j ) ≥ p(wi + j̸=i wj ).

▶ Letting y j → wj for all j 6= i, we have py i ≥ pwi .

(We have shown that (p, X) is a “quasi-equilibrium”.)
▶ We want to show that if y i ∈ Rn+ , U i (y i ) > U i (wi ), and
y i ≤ M , then py i > pwi .

60 / 61
▶ Suppose that y i ∈ Rn+ , U i (y i ) > U i (wi ), and y i ≤ M .

▶ By the continuity of U i , U i (αy i ) > U i (wi ) for some α < 1.

Then, as we have shown, we must have p(αy i ) ≥ pwi .
▶ Since wi  0 and p ≥ 0, p 6= 0, we have
0 < pwi ≤ α(py i ) < py i .

61 / 61