TOPOLOGY AND ECONOMICS

SONGZI DU

1. Introduction
In the first part of this article we will work through the notions
of smooth manifolds, tangent spaces and differentiation on manifolds
(all in the simple Euclidean setting) to prove an important result in
differential topology, Sard’s Theorem. Our exposition is entirely based
on the first three sections of Milnor’s lovely little book Topology from
the Differentiable Viewpoint.
In the second part of this article, we will study a simple but neverthe­
less insightful model of economy with tools of topology. In particular,
we will put the two powerful theorems of topology, Brouwer’s Fixed
Point Theorem and Sard’s Theorem, into attractive uses.

2. Differential Topology in Euclidean Space

2.1. Smooth Map and Manifolds.
Definition 2.1.1. Let U be an open subset in Rk , and let Y be an
arbitrary subset of Rl . The map f : U → Y is smooth if at every point
in U partial derivatives of f of all order exist and are continuous.
Definition 2.1.2. Let X and Y be arbitrary subsets of Rk and Rl ,
respectively. The map f : X → Y is smooth if at every point x ∈ X,
there exist an open set U ⊂ Rk containing x and a smooth extension
(smooth in the sense of Definition 2.1.1) F : U → Rl such that F agrees
with f in U ∩ X.
It is clear that Definition 2.1.2 is consistent with Definition 2.1.1.
We first note that smooth maps are continuous. Suppose the map­
ping f : X → Y is smooth; take any x ∈ X, the smooth extension of
f at x, F : U → Rl is continuous by construction; but then f agrees
with a continuous function (F ) on a neighborhood (U ∩ X) of x in X,
thus f must be continuous at x.
The smoothness of a map is also preserved by restriction to subsets.
Let f defined above be again smooth, and let G be an arbitrary subset

1
2 Songzi Du

of X. Then f |G is smooth: for any x ∈ G, let F : U → Rl be

a smooth extension of f at x; we see immediately that F is also a
smooth extension of f |G at x.
The smoothness of maps is also preserved by composition:
Proposition 2.1.1. Suppose X ⊂ Rk , Y ⊂ Rl , Z ⊂ Rj , f : X → Y
and g : Y → Z are smooth. Then g ◦ f : X → Z is smooth.
Proof. Fix an x ∈ X; let y = f (x) ∈ Y . There exist smooth extensions
F : U → Rl and G : V → Rj where U and V are neighborhoods
(open in Rk and in Rl , respectively) of x and y, respectively. Let
U � = F −1 (V ) ⊂ U ; clearly U � is a neighborhood of x in Rk . Now we
can easily check that G ◦ F |U � : U � → Rj is a smooth extension of g ◦ f
at x. �
Definition 2.1.3. A map f : X → Y is called a diffeomorphism if f
is bijective and if both f and f −1 is smooth.
Clearly, a diffeomorphism is also a homeomorphism. Also, given
G ⊂ X, f |G : G → f (G) is a diffeomorphism if f : X → Y is a
diffeomorphism.
Definition 2.1.4. A subset M ⊂ Rk is called a smooth manifold of
dimension m if each x ∈ M has a neighborhood W in M that is
diffeomorphic to an open subset U of Rm .
Any particular diffeomorphism g : U → W is called a parametriza­
tion of W . The inverse diffeomorphism g −1 : W → U is called a system
of coordinates on W.
It follows from the definition that any discrete set in some Euclidean
space (for example, a finite set) is a smooth manifold of dimension 0
(by definition, R0 = {0}).
It is easily seen that for a manifold M , if N is an open subset of M ,
then N also a manifold (i.e. a submanifold of M ), with dim N = dim M
open subset. Since Rk is trivially a manifold of dimension k, we see
that any of its open subsets is a manifold of dimension k.
2.2. Tangent Spaces and Derivatives.
Definition 2.2.1. For an open subset U of Rk , its tangent space at
any x ∈ U , T Ux , is defined to be the entire vector space Rk . Thus in
this sense manifolds are a generalization of open sets.
Definition 2.2.2. For a smooth map f : U → Y ⊂ Rl , where U is
open in Rk , its derivative at any x ∈ U , dfx : Rk → Rl , is defined as
dfx (h) = limt→0 (f (x + th) − f (x))/t, for h ∈ Rk .
 Topology and Economics 3

From multivariable calculus, we learned that the derivative above

takes the form dfx (h) = Jh, where J is the l × k Jacobian matrix
of partial derivatives. Also from multivariable calculus, we have the
following proposition:
Proposition 2.2.1 (Chain Rule). Let U ⊂ Rk , V ⊂ Rl be open sets
and W ⊂ Rj an arbitrary set, and let f : U → V and g : V → W
be smooth maps. Then for any x ∈ U , d(g ◦ f )x = dgy ◦ dfx , where
y = f (x).
And the following intuitive fact which we have suspected all along
follows easily from the chain rule:
Proposition 2.2.2. If f is a diffeomorphism between open sets U ⊂ Rk
and V ⊂ Rl , then k must equal l, and the linear map dfx : Rk → Rl
must be nonsingular for all x ∈ U .
Proof. Fix an x ∈ U and let y = f (x). The composition f −1 ◦ f is the
identity map on U , thus d(f −1 ◦ f )x = d(f −1 )y ◦ dfx = iRk , where iRk
is the identity map on Rk . Likewise we have dfx ◦ d(f −1 )y = iRl , which
immediately implies our conclusion. �
A partial converse to the above proposition is the following famous
theorem:
Theorem 2.2.3 (Inverse Function Theorem). Let f : U → Rk be a
smooth map, with U open in Rk , and let x in a point in U . If the
derivative dfx : Rk → Rk is nonsingular, then there exists an open set
U � in U containing x such that f maps U � diffeomorphically onto f (U � )
and f (U � ) is open in Rk
Definition 2.2.3. For a smooth manifold M ⊂ Rk of dimension m
and any point x ∈ M , the tangent space of M at x, Tx (M ), is defined
as follows: choose a parametrization g : U → M of a neighborhood
g(U ) of x in M, where U is open in Rm ; then Tx (M ) = dgu (Rm ), where
g(u) = x and the derivative dgu is in the sense of Definition 2.2.2.
The first thing we must do is to show that Definition 2.2.3 is well­
defined:
Proposition 2.2.4. T Mx in Definition 2.2.3 is independent of the
parametrization g : U → M .
Proof. Let h : V → M be another parametrization of another neighbor­
hood h(V ) of x in M (as usual, V is open in Rm ), and let v = h−1 (x).
Our Tx (M ) will be well­defined if dgu (Rm ) = dhv (Rm ).
4 Songzi Du

Let W = g(U ) ∩ h(V ), U � = g −1 (W ) and V � = h−1 (W ). Clearly U �

and V � are open in Rm . Note that h−1 ◦g : U � → V � is a diffeomorphism,
and that we have the following diagram:
h−1 ◦g
U� B / V�
BB g ||
BB h ||
BB ||
B ~||
M
We then take derivatives and apply the chain rule (Proposition 2.2.1)
to get:
d(h−1 ◦g)u
Rm C / Rm
CC dg dhv {{
{
CC u {
CC {{
C! {} {
Rk
Since h−1 ◦ g is a diffeomorphism, d(h−1 ◦ g)u is an isomorphism
by Proposition 2.2.2. Then from the above diagram we conclude that
dgu (Rm ) = dhv (Rm ). �
Proposition 2.2.5. Let M ⊂ Rk be a m­dimensional smooth manifold.
Then for all x ∈ M , Tx (M ) is a m­dimensional vector subspace of Rk .
Proof. Fix an x ∈ M . We see immediately from Definition 2.2.3 that
Tx (M ) is a subspace of Rk . Let g : U → g(U ) ⊂ M be a parametriza­
tion of g(U ) � x, where U is open in Rm and g(U ) open in M ; let
u = g −1 (x). Since the inverse g −1 : g(U ) → U is a smooth function,
there exists a smooth extension F : W → Rm (where W is open in Rk )
that agrees with g −1 on W ∩ g(U ). Let U � = g −1 (W ∩ g(U )); clearly
U � is open in R. Then we have the following diagram:
U� B
inclusion / Rm
BB g { =
BB F {{{
BB {{
B {{
W
We again take derivatives and apply the chain rule to get:
identity
Rm C / Rm
CC dg { =
CC u dFx { {
CC {{
C! {
{{
Rk
The diagram implies that dgu is injective, which means that Tx (M ) =
dgu (Rm ) is of dimension m. �
We now generalize the definition of derivative to smooth maps be­
tween manifolds:
 Topology and Economics 5

Definition 2.2.4. Given two smooth manifolds M ⊂ Rk and N ⊂ Rl ,

of dimension m and n respectively, and a smooth map f : M → N ;
for any x ∈ M let y = f (x) ∈ N . The derivative of f , dfx : Tx (M ) →
Ty (N ), is defined as follows. Let F : W → Rl be the smooth extension
of f at x, where W contains x and is open in Rk . Then for each
v ∈ Tx (M ), define dfx (v) to be equal to dFx (v), where dFx : Rk → Rl
is defined in the sense of Definition 2.2.2.
Proposition 2.2.6. dfx in Definition 2.2.4 maps Tx (M ) into Ty (N )
and is independent of F : W → Rl .
Proof. We first choose parametrization g : U → M ⊂ Rk and h : V →
N ⊂ Rl for neighborhoods g(U ) of x and h(V ) of y; notice that U
is open in Rm and V open in Rn ; let u = g −1 (x) and v = h−1 (y).
Let W � = f −1 (h(V )) ∩ W and U � = f −1 (W � ); since W � is open in M ,
U � ⊂ U is open in Rm ; and clearly, x ∈ g(U � ). For convenience, we
will rename U � to U and g|U � to g. Then we have g(U ) ⊂ W and
f (g(U )) ⊂ h(V ). Thus, we have the following commutative diagram:
WO
F / Rl
O
g h
h−1 ◦f ◦g
U /V
Taking derivatives and applying the chain rule, we have
dFx
RO k / Rl
O
dgu dhv
d(h−1 ◦f ◦g)u
Rm / Rn
Now it is clear that dFx maps Tx (M ) = dgu (Rm ) into Ty (N ) =
dhv (Rn ); thus so does dfx . Additionally, we see that for any w ∈ Tx (M ),
we have dfx (w) = dhv ◦ d(h−1 ◦ f ◦ g)u ◦ (dgu )−1 (w) (simply go around
the bottom of the diagram, note that it does not matter whether or
not dgu is injective), which is independent of the smooth extension
F : W → Rl of f at x. �
We can generalize the chain rule to manifolds as follows:
Proposition 2.2.7 (Chain Rule on Manifolds). Let M ⊂ Rk , N ⊂ Rl ,
P ⊂ Rj be smooth manifolds. If f : M → N and g : N → P are
smooth, with x ∈ M , y = f (x), then d(g ◦ f )x = dgy ◦ dfx .
Proof. First choose a smooth extension G : V → Rj of g at y such that
V � y is open in Rl . Then choose a smooth extension F : U → Rl of
f at x such that U � x is open in Rk and F (U ) ⊂ V (c.f. the proof
of Proposition 2.1.1). Then, for any v ∈ Tx (M ), dfx (v) = dFx (v) and
6 Songzi Du

d(g◦f )x (v) = d(G ◦ F )x (v), while for any w ∈ Ty (N ), dgy (w) = dGy (w).
Our conclusion follows since by the chain rule for open sets we have
d(G ◦ F )x (v) = dGy ◦ dFx (v). �
Proposition 2.2.8. Let M ⊂ N ⊂ Rk where M and N are smooth
manifolds. Then we have Tx (M ) ⊂ Tx (N ) ⊂ Rk for all x ∈ M .
Proof. Let i : M → N be the inclusion map. Then for any x ∈ M ,
dix : Tx (M ) → Tx (N ) is also an inclusion map. �
And we have the following proposition whose proof is completely
analogous to that of Proposition 2.2.2.
Proposition 2.2.9. Let M and N be two smooth manifolds in some
Euclidean spaces. If f : M → N is a diffeomorphism, then for any
x ∈ M with y = f (x), dfx : Tx (M ) → Ty (N ) is an isomorphism of
vector spaces. In particular, the dimension of M must be equal to the
dimension of N .
2.3. Sard’s Theorem.
Theorem 2.3.1 (Sard’s Theorem). Let f : U → Rn be a smooth map,
defined on an open set U ⊂ Rm , and let C = {x ∈ U | rank dfx < n}.
Then, the image f (C) ⊂ Rn has Lebesgue measure zero.
The proof of Sard’s Theorem is rather involved and deserves a treat­
ment in its own section. But let us first explore the various implications
and generalizations of Sard’s Theorem.
Since a set of Lebesgue measure 0 (called a null set ) in Rm cannot
contain any nonempty open set of Rm , the complement of a null set of
Rm is dense in Rm .
Notice that in the statement of Sard’s theorem m < n means that
C = U ; thus in this case the theorem simply says that f (U ) ⊂ is null
in Rm , which is intuitive but nevertheless non­trivial.
Definition 2.3.1. Given a smooth m­dimensional manifold M , a sub­
set R ⊂ M is of measure 0 (or null ) in M if for any x ∈ R and any
parametrization g : U → M of a neighborhood g(U ) of x, g −1 (g(U )∩R)
is of measure 0 in Rm .
Proposition 2.3.2. Suppose that M is a m­dimensional smooth man­
ifold, and that R ⊂ M is null in M , then M − R is dense in M .
Proof. Suppose M − R is not dense in M , then there exists a x ∈
R and V � x open in M such that V ⊂ R. Let g : U → M be
a parametrization of a neighborhood g(U ) of x, where U is open in
Rm . Then g −1 (g(U ) ∩ V ) is open in Rm and is nonempty (it contains
 Topology and Economics 7

g −1 (x)), which implies that it is not null in Rm . But this means that
g −1 (g(U ) ∩ R) which contains g −1 (g(U ) ∩ V ) is also not null in Rm . �
Definition 2.3.2. For a smooth f : M → N from a manifold of
dimension m to a manifold of dimension n, let C be the set of all
x ∈ M such that the derivative at x dfx : Tx (M ) → Tf (x) (N ) has rank
less than n (i.e. not surjective). Then C is called the set of critical
points of f , while f (C ) the set of critical values. Likewise, M − C is
called the set of regular points of f , while N − f (C ) the set of regular
values.
Theorem 2.3.3 (Sard’s Theorem for Manifolds). The set of critical
values of a smooth map f : M → N between manifolds is of measure 0
in N.
Proof. Suppose that N is of dimension n and M of dimension m. Let
K ⊂ N be the set of critical values of f . For any y ∈ K and any
parametrization h : V → N of a neighborhood h(V ) of y, we will show
that h−1 (h(V ) ∩ K ) is null in Rn .
For each x ∈ f −1 (h(V ) ∩ K ) ⊂ M , choose a parametrization gx :
Ux → M of a neighborhood gx (Ux ) of x such that f (gx (Ux )) ⊂ h(V ).
Let Cx ⊂ Ux be the set of critical points of the smooth map h−1 ◦ f ◦gx :
Ux → R n .
Since M is imbedded in some Euclidean space, we can choose a
countable subset I ⊂ f −1 (h(V ) ∩ K ) such that

f −1 (h(V ) ∩ K ) ⊂ gx (Ux )
x∈I
−1
Apply h ◦ f to both sides of the containment above, we get (remember
that K is the image of f over the set of critical points of f ):

h−1 (h(V ) ∩ K ) ⊂ h−1 ◦ f ◦ gx (Ux )

x∈I

Thus for any z ∈ h (h(V ) ∩ K ), z ∈ h−1 ◦ f ◦ gx (Ux ) for some x ∈ I .

−1

But then for that x ∈ I , z ∈ h−1 ◦ f ◦ gx (Cx ). Thus,

h−1 (h(V ) ∩ K ) ⊂ h−1 ◦ f ◦ gx (Cx )

x∈I

Applying Sard’s Theorem to each h−1 ◦ f ◦ gx (x ∈ I ), we conclude

that h−1 (h(V ) ∩ K ) is of measure 0 in Rn . �
Theorem 2.3.4 (Preimage Theorem). If f : M → N is a smooth map
between manifolds of dimensions m ≥ n, and if y ∈ N is a nontrivial
regular value (i.e. f −1 (y) =
� ∅), then the set f −1 (y) ⊂ M is a smooth
manifold of dimension dimension m − n.
8 Songzi Du

Proof. Fix an x ∈ f −1 (y). Since y is a regular value, the rank of

dfx : Tx (M ) → Ty (N ) is n, and the null space ker dfx ⊂ Tx (M ) is an
(m − n)­dimensional vector space.
Suppose that M ⊂ Rk , choose a linear map L : Rk → Rm−n that is
nonsingular on the subspace ker dfx ⊂ Tx (M ). Now define F : M →
N × Rm−n by F (z) = (f (z), L(z)) for z ∈ M . The derivative dFx :
Tx (M ) → Ty (N ) × Rm−n is clearly given by dFx (v) = (dfx (v), L(v)) for
any v ∈ Tx (M ).
Thus, dFx is of rank m. Applying the Inverse Function Theorem
for manifolds, we have F maps U � x open in M diffeomorphically
onto V � (y, L(x)) open in N × Rm−n . Then, F maps f −1 (y) ∩ U
diffeomorphically onto (y × Rm−n ) ∩ V , which is open in y × Rm−n . But
y × Rm−n is diffeomorphic to Rm−n with the natural diffeomorphism
π : y × Rm−n → Rm−n , therefore f −1 (y) ∩ U is the neighborhood of x
in f −1 (y) that is diffeomorphic to the open set π((y × Rm−n ) ∩ V ) of
Rm−n . �

2.4. Proof of Sard’s Theorem. Our proof of Sard’s Theorem follows

§3 of Milnor in an almost verbatim manner. In particular, we will prove
Sard’s Theorem by doing induction on n and m. Note that the theorem
makes sense for m ≥ 0 and n ≥ 1. The theorem is obviously true when
m = 0 for all n ≥ 1.
As in the statement of the theorem, let C be the set of f ’s critical
points. And let Ci ⊂ C be the set of all x ∈ U such that all partial
derivatives of f of order ≤ i vanish at x. Then we have a descending
sequence of sets
C ⊃ C1 ⊃ C2 ⊃ C3 ⊃ . . .
Our proof consists of three steps:

Step 1: f (C − C1 ) has measure zero in Rn . When n = 1, we have

C = C1 , so there is nothing to prove in this step.
We now assume that n ≥ 2.
We will need the following version of Fubini’s Theorem:
Theorem 2.4.1 (Fubini). Let A be a measurable set in Rn . Then A
is of measure 0 in Rn if A ∩ (t × Rn−1 ) is of measure 0 in t × Rn−1 for
all t ∈ R.
For each x̄ ∈ C − C1 we will find a neighborhood V � x open in
Rm such that f (V ∩ C) is of measure 0 in Rn . Since C − C1 can be
covered by countably many of these neighborhoods, this will imply that
f (C − C1 ) is of measure 0 in Rn .
 Topology and Economics 9

Since x¯ �∈ C1 , there exist integers 1 ≤ i ≤ n and 1 ≤ j ≤ m such

that ∂fi /∂xj does not vanish at x¯. Without loss of generosity we can
assume that i = j = 1.
We define the map h : U → Rm by h(x) = (f1 (x), x2 , . . . , xm ) for
all x ∈ U . By construction dhx¯ is nonsingular, thus h maps some
neighborhood V of x¯ in U diffeomorphicaly onto a set V � open in Rm .
Let g = f ◦ h−1 ; then g maps V � into Rn . Note that by the chain rule
the set C � of critical points of g is precisely h(V ∩ C). Thus the set of
critical values of g is f (V ∩ C).
For each (t, x2 , . . . , xm ) ∈ V � , we have g(t, x2 , . . . , xm ) ∈ t × Rn−1 ⊂
Rn , i.e. g maps hyperplanes to hyperplanes. Thus, for a fixed t ∈ R
let πk : t × Rk−1 → Rk−1 be the natural homeomorphic projection
and let Vt� = πm ((t × Rm−1 ) ∩ V � ); notice that Vt� is open in Rm−1
and possibly empty. We derive a map g t : Vt� → Rn−1 from g in the
following manner: for z ∈ Vt� , let g t (z) = πn (g(t × z)).
We can easily check that y ∈ Rn−1 is a critical value of g t if and
only if t × y is a critical value of g. Thus, the πn projection of the
intersection between the set of critical values of g (the set g(C � )) and
t × Rn−1 is equal to the set of critical value of g t . But by our induction
hypothesis applied to g t , the set of critical value of g t is of measure
zero in Rn−1 . Therefore by Fubini’s Theorem we conclude that g(C � )
is of measure 0 in Rn .

Step 2: f (Ci − Ci+1 ) is of measure 0 in Rn for i ≥ 1. For each x̄ ∈

Ci − Ci+1 there is some (i + 1)th derivative ∂ i+1 fr /∂s1 . . . ∂si+1 that
is not zero at x.¯ Let w(x) = ∂ i fr /∂s2 . . . ∂si+1 ; then w(¯
x) = 0 but
∂w/∂xs1 (x̄) �= 0. Without loss of generosity we assume that s1 = 1.
Then the map h : U → Rm defined by h(x) = (w(x), x2 , . . . , xm ) carries
some neighborhood V of x̄ diffeomorphically onto an open set V � of Rm .
Note that h carries Ci ∩V into the hyperplane 0×Rm−1 . Again we let
g = f ◦h−1 : V � → Rn . Let g˜ : πm ((0×Rm−1 )∩V � ) ⊂ Rm−1 → Rn be the
induced map from g analogous to the g t in the previous case. Applying
our induction hypothesis to g̃, we conclude that the set of critical values
of g̃ is of measure 0 in Rn . But πm (h(Ci ∩ V )) are certainly critical
points of g̃, thus f (Ci ∩ V ) = g(h(Ci ∩ V )) = g̃(πm (h(Ci ∩ V ))) is of
measure 0 in Rn . Since Ci − Ci+1 is covered by countably many such
sets V , we conclude that f (Ci − Ci+1 ) has measure 0 in Rn .

Step 3: f (Ck ) is of measure 0 in Rn when k is sufficiently large. Let

I m ⊂ U be a cube with edge δ. It suffices to show that when k is
sufficiently large, f (Ck ∩ I m ) has measure 0 in Rn , since Ck can be
covered by countably many of these Ck ∩ I m .
10 Songzi Du

From Taylor’s theorem, the compactness of I m , and the definition of

Ck , we have
f (x + h) = f (x) + R(x, h)
where
||R(x, h)|| ≤ c||h||k+1
and x ∈ Ck ∩ I m , x + h ∈ I m , and c is a constant that only depends on
f and I m .
We now subdivide I m into rm cubes of edge δ/r. Let I1 be one of
these cubes that contains a point√x ∈ Ck . Then, any point of I1 can
be written as x + h, with ||h|| ≤ mδ/r.
Thus,√f (I1 ) lies in a cube of edge a/rk+1 centered about f (x), where
a = 2c( mδ)k+1 . Hence f (Ck ∩ I m ) is contained in a union of at most
rm cubes having total volume
a
V ≤ rm ( k+1 )n = an rm−(k+1)n
r
Clearly, for k such that k + 1 > m/n, V tends to zero as r tends to
infinity. Therefore, for k > m/n − 1, f (Ck ∩ I m ) is of measure zero in
Rn .

3. Theory of General Equilibrium

General equilibrium is a branch of economic theory that studies the
equilibrium (i.e. when supply equals demands for every goods in the
economy) state of an economy inhabited by agents with different and
sometimes conflicting interests. An important question one asks in
general equilibrium is that for a given model of economy, does an equi­
librium exist? And if so, how many equilibria are there and what
sort of local and global properties do they possess? If this section, we
will attempt to answer these questions for a simple model of economy
where agents consume by exchanging their initially owned goods in a
“competitive” market.
A word on our notation: we let Rn+ = {(x1 , . . . , xn ) | xi ≥ 0 for i =
1, . . . , n} and Rn++ = {(x1 , . . . , xn ) | xi > 0 for i = 1, . . . , n}. For
x, y ∈ Rn , we write x ≥ y when xi ≥ yi for i = 1, . . . , n.
3.1. Model of Exchange Economy. Our economy consists of I con­
sumers (i = 1, . . . , I) and L consumption goods (l = 1, . . . , L). A
bundle of consumption goods is represented by a point in the goods
space RL+ . Each consumption good has a market price, thus the space
of all market prices is also RL+ .
We assume that each consumer i has an initial endowment which
is a bundle of consumer goods represented by ω i ∈ RL . Given the
 Topology and Economics 11

current market prices for consumer goods p ∈ RL , the “net worth”

of this endowment ω i is p · ω i ; this will be the income of consumer i.
Notice that here we have assumed that each consumer is able to sell
all consumer goods in the market according to the current price.
Finally, consumer i might not be entirely happy with his endowment
ω i . Therefore he would like to buy consumer goods in the market
with his income derived from his endowment; that is, he would sell his
endowment in the market, and with the money generated from his sale
he would buy a bundle of goods with which he would be most satisfied.
This behavior is captured by his demand function f i : RL+ × (0, ∞) →
RL+ ; that is, given the current market price p and his income p · ω i , the
demand function gives us a bundle of goods f i (p, p · ωi ) that consumer
i would buy and consume.
We will let the set of all possible endowment bundles of all I con­
sumers to be the space (RL+ )I = RIL IL i
+ . Then for each ω ∈ R+ , ω denotes
the endowment bundle of consumer i, i.e. (ω(i−1)L+1 , . . . , ωiL ) ∈ RL++ ,
which is consistent with what we used before.
Definition 3.1.1. A state of economy characterized by ω ∈ RIL
+ and
p ∈ RL+ is in general equilibrium if
I
� I
�
i i
f (p, p · ω ) = ωi
i=1 i=1

We usually fix an ω ∈ RIL + and look at the set of price vectors that

induce general equilibrium.
The condition general equilibrium simply says that the total demand
of each good is equal to its total supply. Intuitively, a state of econ­
omy not in general equilibrium seems unstable, since some people have
unfulfilled desire, and should converge (via adjusting the price vector
p ∈ RL+ ) to a general equilibrium state. This intuition thus justifies our
focus on general equilibrium.
Finally, we define the aggregate excess demand function Z : RL+ ×
RIL
+ → R by
L

�I
Z(p, ω) = (f i (p, p · ω i ) − ω i )
i=1

Clearly, an economy is in general equilibrium if and only if Z(p, ω) =

0.
For the existence of an equilibrium price vector, we need the follow­
ing assumptions on the demand functions and their derived aggregate
excess demand:
12 Songzi Du

Assumption 3.1. Each f i : RL+ × R+ → RL+ (i = 1, . . . , I) satisfies:

(1) f i is continuous.
(2) f i (cp, cw) = f i (p, w) for all c ∈ (0, ∞), p ∈ RL+ and w ∈ R+ .
(3) f i (p, w) · p = w for all p ∈ RL+ and w ∈ R+ .
Z : RL+ × RIL L
+ → R satisfies:
(4) Given an p ∈ RL+ and an ω ∈ RIL + , for l = 1, . . . , L, pl = 0
implies Zl (p, ω) > 0.
Part (1) of Assumption 3.1 is needed for technical (topological) rea­
sons and seems relatively harmless; after all, people’s behaviors usually
do not change radically given a small change in price and income. Part
(2) says that multiplying the price and his income by a constant factor
will not affect agent i’s consuming decision. Part (3) says that each
consumer will use up all of his income for consumption and will not
waste a dime. Part (4) simply says that everyone wants free goods.
These assumptions seem to describe the consumer­behavior of a typi­
cal materialistic individual in a capitalist economy, and thus seem not
too implausible or restrictive. We will show in the next subsection that
although each person in our model economy is “greedy” and acts only
in self­interest (as dictated by his demand function), there still exists
a set of prices that leaves no desire unfulfilled and no resource wasted.

3.2. Existence of Price Equilibrium. Throughout this subsection

we assume that Assumption 3.1 on the demand functions holds.
Lemma 3.2.1. For any ω ∈ RIL L
+ and any p ∈ R+ , if Z(p, ω) ≤ 0, then
Z(p, ω) = 0.
Proof. Suppose that Zl (p, ω) < 0 for some l. Then pl = 0: since
p ∈ RL+ , pl =
� 0 means Zl (p, ω)pl < 0, which implies that Z(p, ω) · p < 0;
but by Assumption 3.1(3) we have p · Z(p, ω) = 0. However, pl = 0
implies that Zl (p, ω) > 0 by Assumption 3.1(4). �
Lemma 3.2.2. Suppose that ω ∈ RIL L
+ and p ∈ R+ induce general
L
equilibrium, then p ∈ R++ .
Proof. This follows immediately from Assumption 3.1(4). �
We will focus our attention on the price simplex S+L−1 = {p ∈ RL+ |
L−1
p1 + . . . + pL = 1}. Let S++ = S+L−1 ∩ R++
L
.
Theorem 3.2.3 (Existence of General Equilibrium). For any initial
L−1
endowment ω ∈ RIL L
+ , there exists a price vector p ∈ S++ ⊂ R++ that
induces general equilibrium.
 Topology and Economics 13

Proof. Fix an ω ∈ RIL

+ .
We define a map g : S+L−1 → S+L−1 by
pl + max(0, Zl (p, ω))
gl (p) = for l = 1, . . . , L
1 + Ll=1 max(0, Zl (p, ω))
�

Clearly, g is continuous, so Brouwer’s Fixed Point Theorem applies,

and we have g(p∗ ) = p∗ for some p∗ ∈ S+L−1 .
Then, we have for l = 1, . . . , L
L
�
p∗l max(0, Zl (p∗ , ω)) = max(0, Zl (p∗ , ω))
l=1
Multiplying by Zl (p∗ , ω) and summing over l = 1, . . . , L, we arrive
at
� L �� L
� L
� � �
p∗l Zl (p∗ , ω) ∗
max(0, Zl (p , ω)) = Zl (p∗ , ω) max(0, Zl (p∗ , ω))
l=1 l=1 l=1

Since Z(p , ω) · p = 0, we conclude that Z(p∗ , ω) ≤ 0 since each

∗ ∗

term in the summation on the RHS is non­negative. We then apply

Lemma 3.2.1 and Lemma 3.2.2 to p∗ and ω to obtain Z(p∗ , ω) = 0 and
L−1
p∗ ∈ S++ . �
It is clear that without Assumption 3.1(4), we can only show that
Z(p∗ , ω) ≤ 0, i.e. given the price vector p∗ people’s demands are sus­
tainable but there are possibilities of unwanted (and thus free) goods.

3.3. Local Uniqueness of Equilibria. Things would be great if for

a given endowment distribution there exists an unique price vector that
induces general equilibrium. In this case our model makes sharp pre­
diction on the future state of the economy and therefore can be tested
against real economic data. However, only with some unrealistically
strong assumptions could we prove uniqueness of price equilibrium.
Therefore, we will hope for the next best thing: each price equilibrium
is somehow nicely separated from the other equilibria (so that there is
the possibility of some non­equilibrium states of economy converging
to one of them), and they are not too many of them. We will tackle
the first question in this section, using the theory of smooth manifolds
and Sard’s Theorem developed earlier in this article. With additional
tools of differential topology, one can show that the number of price
equilibria is finite and in fact is always odd (see Nagata, §4 and 9).
Although there seems not to be any economic interpretation for this
oddness of equilibria, it does spare us the effort to prove the existence
14 Songzi Du

of equilibrium (i.e. no need to prove Theorem 3.2.3) since zero is not

an odd number!
For a technical reason that will become apparent later, we will as­
sume that both our endowment space and price space are strictly pos­
itive; that is, ω ∈ RIL L
++ and p ∈ R++ . Correspondingly, our demand
functions are now defined only on RL++ × R++ . Additionally, in this
subsection we assume that they satisfy the following conditions:
Assumption 3.2. Each f i : RL++ × R++ → RL+ (i = 1, . . . , I) satisfies:
(1) f i is smooth.
(2) f i (cp, cw) = f i (p, w) for all c ∈ R++ , p ∈ RL++ and w ∈ R++ .
(3) f i (p, w) · p = w for all p ∈ RL++ and w ∈ R++ .
The aggregate excess demand function Z : RL++ × RIL L
++ → R is
still defined as before. For a given ω ∈ RIL ++ , let E(ω) = {p ∈
L−1
S++ | Z(p, ω) = 0}, i.e. the set of price equilibria associated with
L−1
ω. We restrict our attention to S++ because by Assumption 3.2(2) if
Z(p, ω) = 0 then Z(cp, ω) = 0; but p and cp are really be the same
equilibrium, since they differ only in “monetary units.”
Definition 3.3.1. Equilibria in E(ω) are locally unique if E(ω) is dis­
L−1
crete in S++ .
L−1 (I−1)L
We define the map F : S++ × R++ × R++ → RIL by
I
� I
�
F (p, w1 , ω 2 , . . . , ω I ) = (f 1 (p, w1 ) + f i (p, p · ω i ) − ωi, ω2, . . . , ωI )
i=2 i=2

Note that F is a smooth map between two manifolds of dimension

LI.
−1
Proposition 3.3.1. Given an ω ∈ RIL ++ , π maps F (ω) homeomor­
phically onto E(ω), where π is the projection of the first L coordinates.
−1
Proof. Fixed an ω ∈ RIL 1 2 I
++ . Suppose (p, w , ω , . . . , ω ) ∈ F (ω) for
L−1 1
some p ∈ S++ and w ∈ R++ , then by the construction of F we have
I
� I
�
f 1 (p, w1 ) + f i (p, p · ω i ) − ωi = ω1
i=2 i=2

Multiplying by p, using Assumption 3.2(3) and doing some subtractions

we arrive at w1 = p · ω 1 . Plug this back to the equation above, we
conclude that p ∈ E(ω).
Now suppose that we have p ∈ E(ω). Let w1 = p · ω 1 , then we have
(p, w1 , ω 2 , . . . , ω I ) ∈ F −1 (ω). �
 Topology and Economics 15

Thus, to see if E(ω) is locally unique, we only need to check if F −1 (ω)

L−1 (I−1)L
is discrete in S++ × R++ × R++ ; this is when the machinery devel­
oped in the first part of this article becomes handy:
Theorem 3.3.2 (Local Uniqueness of Price Equilibria). The set of
ω ∈ RIL IL
++ such that E(ω) is locally unique is dense in R++ .

Proof. Let ω ∈ RIL be a regular value of F . Then, F −1 (ω) is either an

zero­dimensional manifold or empty by the Pre­image Theorem. For
L−1 (I−1)L
either case F −1 (ω) is discrete in S++ × R++ × R++ .
By Sard’s Theorem, the set of regular values ω ∈ RIL of F is dense
in RIL . Thus, its intersection with RIL IL
++ is dense in R++ . �
3.4. Conclusion. If our demand functions satisfy Assumptions 3.1,
and their restrictions to RL++ × R++ also satisfy Assumption 3.2, then
for almost every initial endowment distribution ω ∈ RIL ++ , its set of
price equilibria E(ω) is both non­empty (i.e. an equilibrium exists)
and locally unique (i.e. equilibria are nicely separated, topologically).
References
[1] John Milnor. Topology from the Differentiable Viewpoint. 1965.
[2] Ryo Nagata. Theory of Regular Economies. 2004.