Etric Spaces Topology and Continuity: U B C E 526

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

M ETRIC SPACES , TOPOLOGY, AND CONTINUITY

PAUL S CHRIMPF
O CTOBER 24, 2016
U NIVERSITY OF B RITISH C OLUMBIA
E CONOMICS 526

Always consider a problem under the minimum structure in which it makes


sense. ... one is naturally led to the study of problems with a kind of minimal and intrinsic structure. Besides the fact that it is much easier to find the
crux of the matter in a simple structure than in a complicated one, there are
not so many really basic structures, so one can hope that they will remain of
interest for a very long time. Talagrand (2005)
This lecture focuses on metric spaces, topology, and continuity. Similar material is covered in chapters 12 and 29, of Simon and Blume (1994), or 1.3 of Carter (2001).
1. S EQUENCES AND LIMITS
A sequence is a list of elements, { x1 , x2 , ...} or { xn }
n=1 or sometimes just { xn }. Although the notation for a sequence is similar to the notation for a set, they should not be
confused. Sequences are different from sets in that the order of elements in a sequence
matters, and the same element can appear many times in a sequence. Some examples of
sequences with xi R include
(1) {1, 1, 2, 3, 5, 8, ...}
(2) {1, 12 , 13 , 14 , ...}
(3) { 12 , 32 , 34 , 54 , 56 , ...}
Some sequences, like 2, have elements that all get closer and closer to some fixed point.
We say that these types of sequences converge. A sequence that does not converge diverges. Some divergent sequences like 1, increase without bound. Other divergent sequences, like 3, are bounded, but they do not converge to any single point.
To analyze sequences with elements that are not necessarily real numbers, we need to
be able to say how far apart the entries in the sequence are.
Definition 1.1. A metric space is a set, X, and function d : X X R called a metric (or
distance) such that x, y, z X
(1) d( x, y) > 0 unless x = y and then d( x, x ) = 0
(2) (symmetry) d( x, y) = d(y, x )
(3) (triangle inequality) d( x, y) d( x, z) + d(z, y).
Example 1.1. R is a metric space with d( x, y) = | x y|.
1

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Example 1.2. Any normed vector spacea is a metric space with d( x, y) = x y.


aIn the past, we covered vector spaces before metric spaces, so this example made more sense here.

Now it can be safely skipped.


n
The most common metric space
n that we will encounter will be R with the Euclidean
metric, d( x, y) = x y = i=1 ( xi yi )2 .

Definition 1.2. A sequence { xn }


n=1 in a metric space converges to x if > 0 N such
that
d( xn , x ) <
for all n N. We call x the limit of { xn }
n=1 and write limn xn = x or xn x.
Example 1.3. The sequence {1, 12 , 13 , 14 , ...} = {1/n}
n=1 converges. To see this, take
any > 0. Then N such that 1/N < . For all n N, d(1/n, 0) = 1/n < .
If a sequence does not converge, it diverges.
Example 1.4. The Fibonacci sequence, {1, 1, 2, 3, 5, 8, ...} diverges.
Definition 1.3. a is an accumulation point of { xn }
n=1 if > 0 infinitely many xi such
that
d( a, xi ) < .
Example 1.5. The sequence { 12 , 32 , 34 , 54 , 56 , ...} has two accumulation points, 1 and
1.
The limit of any convergent sequence is an accumulation point of the sequence. In fact, it
is the only accumulation point.
Lemma 1.1. If xn x, then x is the only accumulation point of { xn }
n =1 .
Proof. Let > 0 be given. By the definition of convergence, N such that
d( xn , x ) <
for all n N. {n N : n N } is infinite, so x is an accumulation point.
Suppose x is another accumulation point. Then > 0 N and N such that if n
N and n N , then d( xn , x ) < /2 and d( xn , x ) < /2). By the triangle inequality,
d( x, x ) d( xn , x ) + d( xn , x ) < . Since this inequality holds for any , it must be that
d( x, x ) = 0. d is a metric, so then x = x , and the limit of sequence is the sequences
unique accumulation point.

The third example of a sequence at the start of this section, 3, shows that the converse
of this lemma is false. Not every accumulation point is a limit.
Definition 1.4. Given { xn }
n=1 and any sequence of positive integers, { nk } such that n1 <
n2 < ... we call { xnk } a subsequence of { xn }
n =1 .
2

METRIC SPACES, TOPOLOGY, AND CONTINUITY

In example 3, there are two accumulation points, 1 and 1, and you can find subsequences that converge to these points.
Lemma 1.2. Let a be an accumulation point of { xn }. Then a subsequence that converges to a.
Proof. We can construct a subsequence as follows. Let {k } be a sequence that converges
to zero with k > 0k, (for example, k = 1/k). By the definition of accumulation point,
for each k infinitely many xn such that
d ( x n , a ) < k

(1)

Pick any xn1 such that (1) holds for 1 . For k > 1, pick nk = n j for all j < k and such that
(1) holds for k . Such an nk always exists because there are infinite xn that satisfy (1). By
construction, limk xnk = a (you should verify this using the definition of limit).

Convergence of sequences is often preserved by arithmetic operations, as in the following two theorems.
Theorem 1.1. Let { xn } and {yn } be sequences in Rn (or any normed vector space) V. If xn x
and yn y, then
xn + yn x + y.
Proof. Let > 0 be given. Then Nx such that for all n Nx ,
d( xn , x ) < /2,
and Ny such that for all n Ny ,
d(yn , y) < /2.
Let N = max{ Nx , Ny }. Then for all n N,
d( xn + yn , x + y) = ( xn + yn ) ( x + y) xn x + yn y

</2 + /2 = .

Theorem 1.2. Let { xn } be a sequence in a normed vector space with scalar field R and let {cn }
be a sequence in R. If xn x and cn c then
xn cn xc.

Proof. Left as an exercise.

In fact, later we will see that if f (, ) is continuous, then lim f ( xn , yn ) = f ( x, y). The
previous two theorems are examples of this with f ( x, y) = x + y and f (c, x ) = cx, respectively.
3

METRIC SPACES, TOPOLOGY, AND CONTINUITY

1.1. Series. Infinite sums or series are formally defined as the limit of the sequence of
partial sums.
n
Definition 1.5. Let { xn }
n=1 be a sequence in a normed vector space. Let sn = i =1 xi
denote the sum of the first n elements of the sequence. We call sn the nth partial sum. We
define the sum of all the xi s as

sn
xi nlim

i =1

This is called a(n infinite) series.


Example 1.6. Let R. i=0 i is called a geometric series. Geometric series appear
often in economics, where will be the subjective discount factor or perhaps 1/(1 +
r ). Notice that
s n =1 + + 2 + + n

= 1 + (1 + + + n 1 )
= 1 + (1 + + + n 1 + n ) n +1
s n (1 ) = 1 n +1
sn =

1 n +1
,
1

so,

i = lim sn

i =0

= lim
=

1 n +1
1

1
if | | < 1.
1

1.2. Cauchy sequences. We have defined convergent sequences as ones whose entries
all get close to a fixed limit point. This means that all the entries of the sequence are also
getting closer together. You might imagine a sequence where the entries get close together
without necessarily reaching a fixed limit.
Definition 1.6. A sequence { xn }
n=1 is a Cauchy sequence if for any > 0 N such that
for all i, j N, d( xi , x j ) < .
It turns that in Rn Cauchy sequences and convergent sequences are the same. This is a
consequence of the way the real numbers are defined.
1.1. Every Cauchy sequence in R converges.
There is a more detailed discussion of this in Chapter 29.1 of Simon and Blume. If you
want more practice with the sort of proofs in this lecture, it would be good to read that
4

METRIC SPACES, TOPOLOGY, AND CONTINUITY

section. The convergence of Cauchy sequences in the real numbers is equivalent to the
least upper bound property that is discussed in the appendix to the lecture notes on sets.
Cauchy sequences do not converge in all metric spaces. For example, the rational numbers are a metric space, and any sequence of rationals that converges to an irrational
number in R is a Cauchy sequence in Q but has no limit in Q. Having Cauchy sequences
converge is necessary for proving many theorems, so we have a special name for metric
spaces where Cauchy sequences converge.
Definition 1.7. A metric space, X, is complete if every Cauchy sequence of points in X
converges in X.
Example 1.7. Rn is a complete metric space. A brief argument follows. You may
want to state the details as an exercise. Let {xi } be a Cauchy sequence in Rn . Each
coordinate of xi = ( x1i , ..., xni ) is a Cauchy sequence in R. R is complete, so each
coordinate has a limit, x ji x j for j = 1, ..., n. Finally, show that x = ( x1 , ..., xn ) is the
limit of the original sequence in Rn .
Example 1.8. p = {( x1 , x2 , ...) s.t. xi R, i=1 | xi | p < } with metric
(
)1/p

d p ( x, y) =

| xi yi | p

i =1

is a complete metric space.


Showing that p is complete is slightly tricky because you have deal with a sequence
of xi p , each element of which is itself an infinite sequence. You should not worry
if you have difficulty following the rest of this example.
To show that p is complete, let {xn }
n=1 be a Cauchy sequence. Let x y p denote
d p ( x, y).
Denote the elements of xi by xi1 , xi2 , .... First, lets show that for any n, x1n , x2n , ...
is a Cauchy sequence in R. Let > 0. Since {xn }
n=1 is Cauchy, N such that for all
i, j N ,


xi x j < .
p
Since



xi x j p =
p

m =1



xim x jm p



All terms in the sum on the right are non-negative and the sum includes xin x jn ,
so




xin x jn p xi x j p
p



xin x jn xi x j
p


Therefore, xin x jn < for all i, j N , i.e. x1n , x2n , ... is a Cauchy sequence in R. R
is complete, so it has some limit. Denote the limit by xn .
5

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Now we will show that x = ( x1 , x2 , ...) is the limit of {xn }


n=1 . First, we should

p
show that x . Let
m

sm =

|xn | p .

n =1

lim sm

We
exists. Since {xn }
n=1 is Cauchy, j such that if i j,
need to show that
xi x j < 1. Using the triangle inequality,



xi xi x j + x j = 1 + x j M
for all i j and some fixed j. Thus, xi M for some constant M and all i j.
Then,
sm =

|xin | p M p

i n =1

n =1

sm

| xn | p = lim

for all m.
is a bounded weakly increasing sequence in R, so it must converge.a

Finally, we should show that {xn }


n=1 converges to x . Let > 0. Since the original
sequence is Cauchy, there is an N such that if i, j > N, then
M





xim x jm p xi x j p <

m =1

for all M. Therefore,


M

lim

j m =1



xim x jm p =

| xim xm
|<

m =1

for all i N and all M. Thus,

xi x = lim

| xim xm
|<

m =1

for all i N, so the sequence converges.


aLet { x } R and suppose x x x ... and { x } is bounded, then we will show { x }
n n =1
n n =1
n n =1
2
3
1

converges. Suppose not. Then the sequence has no accumulation points. In particular, xi is not an
accumulation point of the sequence for any i i.e. there is an > 0 such that for all i there are finitely
many j with d( xi , x j ) < . Then we can construct a subsequence by choosing jk such that jk > jk1 and
| x jk x jk1 | > . But then
x jk = x j1 + ( x j2 x j1 ) + ( x j3 x j2 ) + ... + ( x jk x jk1 )

x j1 + (k 1)
which is not bounded.

2. O PEN SETS
Definition 2.1. Let X be a metric space and x X. A neighborhood of x is the set
N ( x ) = {y X : d( x, y) < .
A neighborhood is also called an open -ball of x and written B ( x ).
6

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Definition 2.2. A set, S X is open if x S, > 0 such that


N ( x ) S.
For every point in an open set, you can find a small neighborhood around that point
such that the neighborhood lies entirely within the set.
Example 2.1. Any open interval, ( a, b) = { x R : a < x < b}, is an open set.
Example 2.2. Any linear subspace of dimension k < n in Rn is not open.
Theorem 2.1.
(1) Any union of open sets is open. (finite or infinite)
(2) The finite intersection of open sets is open.
Proof. Let S j , j J be a collection of open sets. Pick any j0 J. If x j J S j , then there
must be j0 > 0 such that Nj ( x ) S j0 . It is immediate that Nj ( x ) j J S j as well.
0
0
Let S1 , .., Sk be a finite collection of open sets. For each i i > 0 such that Ni ( x ) Si .
Let = mini{1,...,k} i . Then > 0 since it is the minimum of a finite set of positive
numbers. Also, N ( x ) Si for each i, so N ( x ) ik=1 Si .

Definition 2.3. The interior of a set A is the union of all open sets contained in A. It is
denoted as int( A).
From the previous, theorem, we know that the interior of any set is open.
Example 2.3. Here some examples of the interior of sets in R.
(1) A = ( a, b), int( A) = ( a, b).
(2) A = [ a, b], int( A) = ( a, b).
(3) A = {1, 2, 3, 4, ...}, A =
Exercise 2.1. Let X be a metric space and { xn }
n=1 a sequence in X. Show that xn x
if and only if for every open set U containing x N such that xn U for all n N.
3. C LOSED SETS
A closed set is like the opposite of an open set1.
Definition 3.1. A set S X is closed if its complement, Sc , is open.
Theorem 3.1.
(1) The intersection of any collection of closed sets is closed.
(2) The union of any finite collection of closed sets is closed.
(
)c
Proof. Let Cj , j J be a collection of closed sets. Then j J Cj = j J Cjc . Cjc are open, so
(
)c
by theorem 2.1, j J Cjc = j J Cj = is open.

The proof of part 2 is similar.


1However, a set can be both open and closed, and a set can be neither, so they are not exactly opposites.

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Example 3.1 (Closed sets). Some examples of closed sets include


(1) [ a, b] R
(2) Any linear subspace of Rn
(3) {( x, y) R2 : x2 + y2 1}
Closed sets can also be defined as sets that contain the limit of any convergent sequence
in the set. Simon and Blume use this definition. The next theorem shows that their definition is equivalent to ours.
Theorem 3.2. Let { xn } be any convergent sequence with each element contained in a set C. Then
lim xn = x C for all such { xn } if and only if C is closed.
Proof. First, we will show that any set that contains the limit points of all its sequences is
closed. Let x C c . Consider N1/n ( x ). If for any n, N1/n ( x ) C c , then C c is open, and C is
closed as desired. If for all n, N1/n ( x ) C c , then yn N1/n ( x ) C. The sequence {yn }
is in C and yn x. However, by assumption C contains the limit of any sequence within
it. Therefore, there can be no such x, and C c must be open and C is closed.
Suppose C is closed. Then C c is open. Let { xn } be in C and xn x. Then d( xn , x )0,
and for any > 0, xn N ( x ). Hence, there can be no neighborhood of x contained in
C c . C c is open by assumption, so x C c and it must be that x C.

Definition 3.2. The closure of a set S, denoted by S (or cl(S)), is the intersection of all
closed sets containing S.
Example 3.2. If S is closed, S = S.
Example 3.3. (0, 1] = [0, 1]
Lemma 3.1. S is the set of limits of convergent sequences in S.
Proof. Let { xn } be a convergent sequence in S with limit x. If C is any closed set containing
S, then { xn } is in C and by theorem 3.2, x C. Therefore, x S.
Let x S. For any > 0, N ( x ) S = because otherwise N ( x )c is a closed set
containing S, but not x. Therefore, we can construct a sequence xn S N1/n ( x ) that
converges to x and is in S.

Example 3.4. {1/n}nN = {0, 1, 1/2, 1/3, ...}


Definition 3.3. The boundary of a set S is S Sc .
Example 3.5. The boundary of [0, 1] is {0, 1}.

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Example 3.6. The boundary of the unit ball, { x R2 : x < 1} is the unit circle,
{ x R2 : x = 1 } .
Lemma 3.2. If x is in the boundary of S then > 0, N ( x ) S = and N ( x ) Sc = .
Proof. As in the proof of lemma 3.1, all -neighborhoods of x S must intersect with S.
The same applies to Sc .

Exercise 2.1 and theorem 3.2 show that there is an important relationship between convergence of sequences and open and closed sets. Given a definition of what it means for
a sequence to converge, we could use theorem 3.2 to define closed sets. Open sets are
then defined as the complement of closed sets. Conversely, if we specify which sets are
open and closed, we can then define convergence of sequences as in exercise 2.1. Metrics,
convergence of sequences, and open and closed sets are three different ways of describing
the continuity properties of a set.
4. C OMPACT SETS
Compact sets are a generalization of finite sets. Compact sets are essential for proving many important theorems. Compact sets have a somewhat difficult to understand
definition, but they are incredibly useful.
Definition 4.1. An open cover of a set S is a collection of open sets, { G } A such that
S A G .
Example 4.1. Some open covers of R are:
{R}
{(, 1), (1, )
{..., (3, 1), (2, 0), (1, 1), (0, 2), (1, 3), ...}
{( x, y) : x < y}
The first two are finite open covers since they consist of finitely many open sets. The
third is a countably infinite open cover. The fourth is an uncountably infinite open
cover.
Example 4.2. Let X be a metric space and A X. The set of open balls of radius centered at all points in A is an open cover of A. If A is finite / countable / uncountable,
then this open cover will also be finite / countable / uncountable.
Open covers of the form in the previous example are often used to prove some property
applies to all of A by verifying the property in each small N ( x ). Unfortunately, this often
involves taking a maximum or sum of something for each set in the open cover. When
the open cover is infinite, it can be hard to ensure that the infinite sum or maximum stays
finite. When we have a finite open cover, we know that things will remain finite.
Definition 4.2. A set K is compact if every open cover of K has a finite subcover.
9

METRIC SPACES, TOPOLOGY, AND CONTINUITY

By a finite subcover, we mean that there is finite set G1 , ...Gk such that S kj=1 G j .
Compact sets are a generalization of finite sets. Many facts that are obviously true of
finite sets are also true for compact sets, but not true for infinite sets that are not compact.
Suppose we want to show a set has some property. If the set is compact, we can cover it
with a finite number of small balls and then we just need to show that each small ball
has the property we want. We will see many concrete examples of this technique in the
next few weeks.
Example 4.3. R is not compact. {..., (3, 1), (2, 0), (1, 1), (0, 2), (1, 3), ...} is an
infinite cover, but if we leave out any single interval (the one beginning with n) we
will fail to cover some number (n + 1).
Example 4.4. Let K = { x }, a set of a single point. Then K is compact. Let { G }A be
an open cover of K. Then such that x G . This single set is a finite subcover.
Example 4.5. Let K = { x1 , ..., xn } be a finite set. Then K is compact. Let { G }A be
an open cover of K. Then for each i, i such that xi Gi . The collection { G1 , ...Gn }
is a finite subcover.
Example 4.6. (0, 1) R is not compact. {(1/n, 1)}
n=2 is an open cover, but there
can be no finite subcover. Any finite subcover would have a largest n and could not
contain, e.g. 1/(n + 1).
Example 4.7. Let x V, a normed vector space. Let K = { x 12 , x 23 , x 43 , ...}. Then K is
n
not compact. Consider the open cover N x 1 ( x n+
1 ) for n = 1, 2, .... Assuming
3( n +2)2

x = 0, each of these neighborhoods contains exactly one point of K, so there is no


finite subcover.
Before using compactness, lets investigate how being compact relates to other properties of sets, such as closed/open.
Lemma 4.1. Let X be a metric space and K X. If K is compact, then K is closed.
Proof. Let x K c . The collection { Nd( x.y)/3 (y)}, y K is an open cover of K. K is compact,
so there is a finite subcover, Nd( x,y1 )/3 (y1 ), ..., Nd( x,yn )/3 (yn ). For each i, Nd( x,yi )/3 (yi )
Nd( x,yi )/3 ( x ) = , so

in=1 Nd(x,yi )/3 ( x )


is an open neighborhood of x that is contained in K c . K c is open, so K is closed.

Lemma 4.2. Let X be a metric space, C K X. If K is compact and C is closed. Then C is also
compact.
Proof. Let { G }A be an open cover for C. Then { G }A plus C c is an open cover for K.
Since K is compact there is a finite subcover. Since C K, the finite subcover also covers
C. Therefore, C is compact.

10

METRIC SPACES, TOPOLOGY, AND CONTINUITY

An equivalent definition of compactness is in terms of collection of sets with the finite


intersection property. A collection of sets {C }B has the finite intersection property if
for all finite subsets, F A, the intersection, F C is not empty.
Lemma 4.3. Let X be a metric space and K X. K is compact if and only if for every collection
of closed subsets, {C }B with C K, with the finite intersection property, the intersection of
all the subsets is not empty i.e. B C is not empty.
Proof. Left as an exercise. The statement in the lemma is basically the contrapositive of
the definition of compact, combined with the observation that the complement of closed
sets are open and vice versa, and the fact that ( A B)c = Ac Bc .

The definition of compactness is somewhat abstract. We just saw that compact sets are
always closed. Another property of compact sets is that they are bounded.
Definition 4.3. Let X be a metric space and S X. S is bounded if x0 S and r R
such that
d( x, x0 ) < r
for all x S.
A bounded set is one that fits inside an open ball of finite radius. For subsets of R this
definition is equivalent to there being a lower and upper bound for the set. For subsets of
a normed vector space, if S is bounded then there exists some M such that x < M for
all x S.
Lemma 4.4. Let K X be compact. Then K is bounded.
Proof. Pick x0 K. { Nr ( x0 )}rR is an open cover of K, so there must be a finite subcover.
The finite subcover has some maximum r . Then K Nr ( x0 ), so K is bounded.

This lemma along with lemma 4.1 show that if a set is compact then it is also closed and
bounded. In Rn , the converse is also true.
Theorem 4.1 (Heine-Borel). A set S Rn is compact if and only if it is closed and bounded.
Proof. We already showed that if S is compact, then it is closed and bounded.
Now suppose S is closed and bounded. Since S is bounded, it is a subset of some ndimensional cube, say [ a, a]n (i.e. the set of all vectors x = ( x1 , .., xn ) with a xi a).
We will show [ a, a]n is compact, and then use the fact that a closed subset of a compact
set is compact.
Lets just show [ a, a]n is compact for n = 1. The argument for larger n is similar, but
the notation is more cumbersome. If [ a, a] is not compact, then there is an infinite open
cover with no finite subcover, say { G }A . If we cut the interval into two halves, [ a, 0]
and [0, a], at least one of them must have no finite subcover. We can repeat this argument
many times to get nested closed intervals of length a/(2k ) for any k. Call the kth interval
Ik . We claim that
k=1 Ik = . To show this take the sequence of lower endpoints of the
intervals, call it { xn }
n=1 . This is a Cauchy sequence, so it converges to some limit, x0 .

Also, for any k, { xn }n=k is a sequence in Ik . Ik is closed so x0 Ik . Thus x0


k=1 Ik . On
11

METRIC SPACES, TOPOLOGY, AND CONTINUITY

the other hand, Ik A G for all k. Therefore, x0 must be in some open G as part of
this cover. Then > 0 such that N ( x0 ) G . However, for k > 1/, Ik N ( x0 ) G ,
and then Ik has a finite subcover. Therefore, [ a, a] must be compact.
The argument for n > 1 is very similar. For n = 2, we would divide the square [ a, a]2
into four smaller squares. For n = 3, we would divide the cube into eight smaller cubes.
In general we would divide the hypercube [ a, a]n into 2n hypercubes with half the side
length.

You may wonder whether closed and bounded sets are always compact. We will later
see that all finite dimensional real vector spaces are isomorphic to Rn . In any such space,
sets are compact iff they are closed and bounded. However, in infinite dimensional
spaces, there are closed and bounded sets that are not compact. The argument in the
previous proof does not apply to infinite-dimensional spaces because an infinite dimensional hypercube can only be divided into infinitely many hypercubes with half the side
length.
Example 4.8. = {( x1 , x2 , ...) : supi | xi | < with norm x = supi | xi | is a normed
vector space. Let ei be the element of all 0s except for the ith position, which is 1. Then
E = {ei }i=1 is closed and bounded. However, E is not compact because { N1/2 (ei )}i=1
is an open cover with no finite subcover.
It is always true that a closed subset of a compact set is compact.
Lemma 4.5. Let C K X, where X is a metric space. If K is compact and C is closed, then C
is compact.
Proof. Let { G }A be an open cover for C. C is closed, so C c is open. Also, C c along with
{ G }A is an open cover of K. K is compact so there exists a finite subcover. This finite
subcover (with C c removed) is also a finite subcover of C. Therefore, C is compact.

We saw that closed sets contain the limit points of all their convergent sequences. There
is also a relationship between compactness and sequences.
Definition 4.4. Let X be a metric space and K X. K is sequentially compact if every
sequence in K has an accumulation point in K.
Sometimes this definition is written as: K is sequentially compact if every sequence in
K has a subsequence that converges in K. Compactness implies sequential compactness.
Lemma 4.6. Let X be a metric space and K X be compact. Then K is sequentially compact.
Proof. Let { xn }
n=1 be a sequence in K. Pick any > 0, N ( x ), x K is an open cover of
K, so there is a finite subcover. Therefore, one of the neighborhoods must contain an
infinite number of the elements from the sequence. Call this neighborhood N ( x1 ). Pick
the smallest n such that xn N ( x1 ) and call it n1 . N ( x1 ) K is a closed subset of the
compact set K, so is itself compact. Repeat the above argument with /2 in place of
and N ( x1 ) K in place of K to find an n2 , n3 , etc. Then the subsequence xn1 , xn2 , ... is a
Cauchy sequence, so it converges. Its limit is an accumulation point. Its limit must be in
K because K is compact, and so, closed. Therefore, K is sequentially compact.

12

METRIC SPACES, TOPOLOGY, AND CONTINUITY

In Rn , a set is sequentially compact iff it is compact iff it is closed and bounded.


Theorem 4.2 (Bolzano-Weierstrass). A set S Rn is closed and bounded if and only if it is
sequentially compact.
Proof. Let S be closed and bounded. By the Heine-Borel theorem (4.1), S is compact. By
lemma 4.6, S is sequentially compact.
Let S be sequentially compact. Let { xn } be a convergent sequence in S. Its limit is an
accumulation point, so it must be in S. Therefore, S is closed. To show S is bounded, pick
x0 S. Suppose x1 S such that d( x1 , x0 ) 1, and x2 S such that d( x2 , x0 ) 2 etc.
This sequence is not Cauchy because of the reverse triangle inequality,


d ( x i , x j ) d ( x i , x0 ) d ( x j , x0 ) = | i j |
This would be a sequence in S with no accumulation points. Therefore, it must not always
be possible to find such xn . In other words, S must be bounded.

Comment 4.1. This theorem is sometimes stated as each bounded sequence in Rn


has a convergent subsequence. As an exercise, you may want to verify that this
statement is equivalent to the one above.
Simon and Blume also prove this theorem in chapter 29.2. They do not prove the HeineBorel theorem first though, so their proof is of the Bolzano-Weierstrass theorem is longer.
Perhaps unsurprisingly, the details of their proof are somewhat similar to our proof of the
Heine-Borel theorem.
In Rn , compactness, sequential compactness, and closed and bounded are all the same.
In general metric spaces, this need not be true. We saw above that in , there are closed
and bounded sets which are not compact. However it is always true that sequential compactness and compactness are the same for metric spaces. We already showed that compactness implies sequential compactness. The proof that sequential compactness implies
compactness is a somewhat long and difficult, and you may want to skip it unless you
are especially interested.
Theorem 4.3. Let X be a metric space and K X. K is compact if and only if K is sequentially
compact.
Proof. Lemma 4.6 shows that if K is compact, then K is sequentially compact.
Suppose every sequence in K has a convergent subsequence with a limit point in K. Let
G , A be an open cover of K. A could be uncountable, so we will begin by showing
that there must be a countable subcover. Let n = 1. Pick x1 K. If possible choose x2 K
such that d( x1 , x2 ) 1/n. Repeat this process, choosing x j in K such that d( x j , xi ) 1/n
for each i < j. Eventually this will no longer be possible because otherwise we could
construct a sequence with no convergent subsequence. When it is no longer possible, set
n = n + 1. This gives a countable collection of open neighborhoods N1/n ( xi ) that cover
K for each n and get arbitrarily small as n increases. Call these neighborhoods j for
j = 1, 2, ... Let J be set of all j such that j G for some . J is a subset of a countable
set, so J is countable. Note that j J j K because if x K, then x G for some , and
13

METRIC SPACES, TOPOLOGY, AND CONTINUITY

then such that N ( x ) G and j s.t. j N ( x ). Finally, for each j J choose G j


such that j G j . Such j exist by construction. Also j J G j j J j K. So G j is a
countable subcover.
If G j has no finite subcover, then for each n,
Fn = (in=1 Gi )c K
is not empty (if it were empty, then in=1 Gi would be a finite subcover). Choose xn Fn .
Then { xn } is a sequence in K, and it must have a convergent subsequence with a limit, x0 ,
in K. However, each Fi+1 Fi and Fi are all closed. Therefore, the sequence { x j }
j=i is also

in Fi and so is its limit. Then x0 i=1 Fi . However,

i=1 Fi = (i=1 Gi )c K,
but Gi is a countable cover of K, which implies

i=1 Fi = (i=1 Gi )c K =
and we have a contradiction. Therefore, G j must have a finite subcover, and K is compact.

Comment 4.2. There are non-metric spaces where sequential compactness and compactness are not equivalent. One can define open sets on a space without a metric
by simply specifying which sets are open and making it such that theorem 2.1 holds.
Such a space is called a topological space. You can then define closed sets, compact
sets, and sequential compactness in terms of open sets. Exercise 2.1 showed that it
is possible to define the convergence of sequences using only open sets, without referring to a metric at all. Similarly, you can define continuity of functions in terms of
open and closed sets. Topology is the branch of mathematics that studies topological spaces. One interesting observation is that on Rn , if a set is open with respect to
some p-norm, then it is also open with respect to any other p-norm. Thus, we say that
Rn with the p-norms are topologically equivalent or homeomorphic. Properties like
continuity and compactness are the same regardless of what p-norm we use.
Topological spaces that are not metric spaces do not come up too often in economics, so we will not study them in detail. An example of a non-metrizable topological
space is the following. Let X = { f : RR}. Define convergence of sequences in this
space as f n f if f n ( x ) f ( x ) for all x R. Let S X be closed if S contains the
accumulation points of all sequences in S. This definition of closed and open sets is
called the topology of pointwise convergence. This space and definition of convergence sounds reasonable (and is reasonable), but there is no metric on this space that
leads to the same definition of convergence and closed sets.
In an area of econometrics and statistics called empirical process theory, you often
have to work with something called the topology of weak convergence, which has
a somewhat similar definition and is also non-metrizable. We saw in example 4.8
that in an infinite dimensional space, the unit sphere (or a closed neighborhood of
any radius) is not compact. It is also not sequentially compact. Econometricians care
14

METRIC SPACES, TOPOLOGY, AND CONTINUITY

about sequential compactness to show that limits and asymptotic distributions exist.
Working with weak convergence instead of a metrizable topology makes many nice
sets sequentially compact.
To review, in Rn a set is compact if any of the following four things hold:
(1) For every open cover there exists a finite subcover,
(2) Every collection of closed sets with the finite intersection property has a nonempty intersection
(3) Every sequence in the set has a convergent subsequence, or
(4) The set is closed and bounded.
The first two are always equivalent. In infinite dimensional spaces, closed and bounded
sets need not be compact, but compact sets are always closed and bounded. In any metric
space, a set is compact iff it is sequentially compact.
5. F UNCTIONS AND CONTINUITY
We have already used functions in this course, so perhaps we should have defined them
earlier. Anyway, a function from a set A to a set B is a rule that assigns to each a A one
and only one b B. If we want to call this function f , we denote this by f : A B, which
is read as f is a function from A to B or simply f from A to B. The set A is called the
domain of f . B is called the target of f . The set

{y B : f ( x ) = y for some x A}
is called the image or range of f .
Example 5.1.
(1) Production functions: f : R2 R
Linear f ( x1 , x2 ) = a1 x1 + a2 x2
Cobb-Douglas: f ( x1 , x2 ) = Kx11 x22
Constant elasticity of substitution: f ( x1 , x2 ) = K (c1 x1a + c2 x2a )b/a
(2) Utility functions: u : RT R
c

Constant relative risk aversion: u(c1 , ..., c T ) = tT=1 t 1t


Constant absolute risk aversion: u(c1 , ..., c T ) = tT=1 t (ect )
(3) Demand function with constant elasticity, D : R3 R2
(
)
Mp111 p212 y 1
D ( p1 , p2 , y ) =
Mp121 p222 y 2
where p1 and p2 are the prices of two goods and y is income.
I do not expect you to remember the names of these functions, but it is very likely that
you will repeatedly encounter them this year.
A continuous function is a function without any jumps or holes. Formally,
Definition 5.1. A function f : X Y where X and Y are metric spaces is continuous at x
if whenever { xn }
n=1 converges to x in X, then f ( xn ) f ( x ) in Y.
15

METRIC SPACES, TOPOLOGY, AND CONTINUITY

We simply say that f is continuous if it is continuous at every x X. There are some


equivalent definitions of continuity that are also useful. You may have seen continuity
defined as in the following lemma.
Lemma 5.1. f : X Y is continuous at x if and only if for every > 0 > 0 such that
d( x, x ) < implies d( f ( x ), f ( x )) < .
Proof. Suppose f is continuous and there is an > 0 such that such that for any > 0,
d( x, x ) < does not imply d( f ( x ), f ( x )) < . Then by letting = 1/n we can construct
a convergent sequence by choosing xn such that d( x, xn ) < 1/n and d( f ( x ), f ( xn )) .
xn x, but f ( xn ) f ( x ). Therefore, if f is continuous it must be impossible to construct
such a sequence. This means that there must be some > 0 such that d( x, x ) < implies
d( f ( x ), f ( x )) < .
Now suppose > 0 > 0 such that d( x, x ) < implies d( f ( x ), f ( x )) < . Let
xn x. Then N s.t. n N implies d( x, xn ) < . But this implies d( f ( x ), f ( xn )) < , so f
is continuous.

Both the definition of continuity and (5.1) are about f being continuous at a point x. We
say that f is continuous on U X if f is continuous at all x U.
A third way of defining continuity is in terms of open sets. First, another definition.
Definition 5.2. Let f : X Y. f is continuous The preimage of V Y is the set in X,
f 1 (V ) defined by
f 1 ( V ) = { x X : f ( x ) V }
A function is continuous if and only if the preimage of any open set is open.
Lemma 5.2. f : X Y is continuous at x if and only if for all open V with f ( x ) V, open
U X such that x U f 1 (V ).
Proof. Suppose for all open V Y that open U such that x U f 1 (V ). We want
to show that then f is continuous on U. To do that, let xn x and let > 0. N ( f ( x )) is
open, so by assumption, open U such that x U f 1 ( N ( f ( x ))). By the definition of
open sets, > 0 such that N ( x ) U f 1 ( N ( f ( x ))). By the definition of xn x, N
such that if n N, xn N ( x ). Then xn f 1 ( N ( f ( x ))), so f ( xn ) N ( f ( x )), i.e.
d ( f ( xn ), f ( x )) < .
Therefore, f ( xn ) f ( x ).
Conversely, suppose f is continuous. Let V Y be open. We want to show that
open U such that x U f 1 (V ). Suppose there is no such U. Then for any > 0,
x f 1 (V ) with
d( x, x ) < .
Pick a sequence of n that converges to zero, such as n = 1/n. Then the associated x n x.
However, since each x n f 1 (V ), f ( x n ) V c . But then having f ( x n ) f ( x ) would mean
that V c is not closed, which contradict V being open. Thus, there must exist an open U
such that x U f 1 (V ) when f is continuous.

16

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Since the a set is open if and only its complement is closed, we can also define continuity
using closed sets.
Corollary 5.1. f : X Y is continuous if and only if f 1 (V ) is closed for all closed V Y.
Proof. Let V Y be closed. Then V c is open. Also, note that the complement of the
preimage of V is the preimage of V c . In symbols,
f 1 ( V ) c = { x X : f ( x ) V } = { x X : f ( x ) V c } = f 1 ( V c ).
From lemma 5.2, f is continuous iff f 1 (V c ) = f 1 (V )c is open for all open sets V c , which
is true iff f 1 (V ) is closed for all closed sets V.

Earlier we saw that convergence of sequences is preserved by arithmetic. Since continuity can be defined using sequences, it should be no surprise that continuity is also
preserved by arithmetic.
Theorem 5.1. Let f : X Y and g : X Y be continuous and X and Y be vector spaces. Then
( f + g)( x ) = f ( x ) + g( x ) is continuous.
Proof. If f and g are continuous, then by definition f ( xn ) f ( x ) and g( xn ) g( x ) whenever xn x. From the previous lecture the limit of a (finite) sum is the sum of limits, so
f ( xn ) + g( xn ) f ( x ) + g( x ), and f + g is continuous.

Similar results can be shown for subtraction, multiplication, etc, whenever they are well
defined.
Continuity is also preserved by composition.
Theorem 5.2. Let f : X Y and g : Y Z be continuous where X, Y, and Z are metric spaces.
Then f g is continuous, where

( f g)( x ) = f ( g( x )).
Proof. Let xn x. g is continuous, so g( xn ) g( x ). f is also continuous, so f ( g( xn )) f ( g( x )).

f g is called the composition of f and g.


5.1. Onto, one-to-one, and inverses. We have already used the concepts of onto, one-toone, and inverses. We restate the definitions here.
Definition 5.3. f : X Y is one-to-one or injective if for all x1 , x2 X,
f ( x1 ) = f ( x2 )
if and only if x1 = x2 .
Equivalently, f is injective if for each y Y, the set { x : f ( x ) = y} is either a singleton
or empty. In terms of a nonlinear equation, if f is one-to-one, then f ( x ) = b has at most
one solution.
Definition 5.4. f : X Y is onto or surjective if y Y, x X such that f ( x ) = y.
17

METRIC SPACES, TOPOLOGY, AND CONTINUITY

In terms of a nonlinear equation, if f is onto, then f ( x ) = b has at least one solution.


When f is one-to-one and onto, we say that f is bijective. A bijective function has an
inverse.
Definition 5.5. If f : X Y is bijective, then the inverse of f , written f 1 satisfies
f ( f 1 (y)) = y
and
f 1 ( f ( x )) = x.
Comment 5.1. While writing these notes, I briefly tried to prove that if f : X Y is
bijective and continuous, then f 1 is continuous. I could not do this, which is good,
because that statement is false. You have to be a little creative in defining X and Y to
come up with a counterexample. Let X = [0, 2 ) and Y = {( x, y) R2 : x2 + y2 = 1}.
Then f ( x ) = (cos( x ), sin( x )) is bijective and continuous, but f 1 is not continuous at
(1, 0).
This counterexample is actually related to a fundamental fact in topology. You may
remember from last lecture that topology is about studying spaces with open and
closed sets that do not necessarily have a metric. One thing that people are interested
when studying such spaces is finding a continuous (in both directions) bijections between them. Loosely speaking, two topological spaces will have a continuous bijection between them if one can be bent and stretched from one into the form of another.
You cannot bend a circle into an interval without breaking the circle, so there is no
continuous bijection between the circle and an interval. When there is a continuous bijection between two spaces, they have the same collection of open sets, so to
a topologist, they are the same. We then call the spaces homeomorphic (or topologically isomorphic). Loosely speaking, spaces will be homeomorphic if they are the
same dimension and their shapes have the same number of holes. The circle has one
hole, an interval has none, so they are not topologically isomorphic. Id be remiss not
to make a joke now, so here goes: Why did the topologist eat her/his coffee mug and
drink from his/her donut? Because theyre topologically isomorphic. Hahaha.
We begin the course by studying how to solve optimization problems. However, we did
not worry too much about conditions that ensure an optimum exists. One fundamental
result that ensures the existence of optima is Weierstrasss theorem.
Theorem 5.3 (Weierstrass). Let X be a metric space and f : X R be continuous and K X
be compact and nonempty. Then x K such that f ( x ) f ( x ) x K.
Proof. First, we show that supxK f ( x ) exists. R has the least upper bound property, so
it is enough to show that f is bounded. Let > 0. For each x, let x Ux be open and
Ux f 1 ( N ( x )). Such Ux exist since f is continuous. Also, {Ux } xK is open cover of K.
K is compact, so there exists a finite subcover. Let this subcover be Ux1 , ..., Uxn . For any
x K, x Ux j for some j and
f ( x ) f ( x j ) + .
18

METRIC SPACES, TOPOLOGY, AND CONTINUITY

Hence,
f (x)

max

j{1,...,n}

f (xj ) +

and f is bounded on K. R has the least upper bound property any set bounded above
has a least upper bound, so f = supxK f ( x ) exists.
Let f = supxK f ( x ). Let yn = f 1/n. By definition of supremum, for each n, xn
such that yn f ( xn ) f. Since K is compact, the sequence { xn } must have a convergent subsequence xnk x K. f is continuous, so f ( xnk ) f ( x ). Also, by construction,
d( f ( xnk ), f)0, so it must be that f ( x ) = f.

6. C ORRESPONDENCES
A function, f : X Y associates exactly one element of Y, f ( x ), with each x X.
Often we encounter things that are like functions, but for each x X, there are multiple
elements of Y. We call this generalization of a function as correspondence.
2

Definition 6.1. A correspondence from a set X to a set Y, is a rule that assigns to each a
Y.
x X a subset of Y. We denote a correspondence by : X

Y is a function from X to the power set of Y.


An equivalent definition is that : X

Correspondences appear often in economics, especially as constraint sets in optimization


problems.
Example 6.1 (Budget correspondence). Suppose there are n goods with prices p Rn .
Then given income of m, a consumer can afford ( p, m) = { x X Rn : p x m},
X. We can write the consumers problem
which defines a correspondence : Rn+1

of maximizing utility subject to the budget constraint as


max u( x )

x ( p,m)

If this problem has a solution, then the indirect utility function is the maximized utility,
v( p, m) = max u( x ).
x ( p,m)

The demand correspondence (usually function) is


x ( p, m) = arg max u( x ).
x ( p,m)

Such maximization problems are central to economics. To derive properties of the


indirect utility and demand functions it is often useful to treat the budge set as a
correspondence.
Correspondences also appear in economics in models with multiple equilibria, such as
many games.
Defining continuity is a bit more complicated for correspondences than for functions.
A function can either be continuous or it can jump. A correspondence can also expand or
2This section is largely based on section 2.1.5 of Carter (2001).

19

METRIC SPACES, TOPOLOGY, AND CONTINUITY

R defined by
contract. For example, consider : R

{
[0, 1]
if x > 0
(x) =
[1/4, 3/4] if x 0
R defined by
and : R

{
[0, 1]
if x 0
( x ) =
[1/4, 3/4] if x < 0
Both these correspondences are somewhat continuous because they contain a continuous
function, e.g. f ( x ) = 1/2, for all x. However, they are also somewhat discontinuous
because the corresponding set changes suddenly at 0. Motivated by this observation we
define the following:

Y is upper hemicontinuous at x if for all seDefinition 6.2. A correspondence, : X

quences xn x and yn ( xn ) with yn y, then y ( x ).


In the previous example, is upper hemicontinuous at 0, but is not. To see this
consider xn = 1/n and yn = 1.

Y is lower hemicontinuous at x if for all seDefinition 6.3. A correspondence, : X

quences xn x and y ( x ), there exists a subsequence, xnk and yk ( xnk ) with yk y.


In the previous example, is lower hemicontinuous at 0, but is not. To see this
consider xn = 1/n and y = 1.
Definition 6.4. We say that a correspondence is continuous if it is both upper and lower
hemicontinuous.
At all x = 0, and are continuous.

Y can also be viewed as a function : X P (Y ).


Exercise 6.1. A correspondence : X

There is a metric on P (Y ) that leads to the same definition of continuity as in definition 6.4. What is that metric?
R EFERENCES
Carter, Michael. 2001. Foundations of mathematical economics. MIT Press.
Simon, Carl P and Lawrence Blume. 1994. Mathematics for economists. Norton New York.
Talagrand, Michel. 2005. The generic chaining. Springer.

20

You might also like