Open and Closed Sets

Definition: A subset S of a metric space (X, d) is open if it contains an open ball about each of
its points i.e., if
x S :  > 0 : B(x, ) S.

(1)

Theorem:
(O1) and X are open sets.
(O2) If S1 , S2 , . . . , Sn are open sets, then ni=1 Si is an open set.
(O3) Let A be an arbitrary set. If S is an open set for each A, then A S is an open set.
In other words, the union of any collection of open sets is open. [Note that A can be any set, not
necessarily, or even typically, a subset of X.]
Proof:
(O1) is open because the condition (1) is vacuously satisfied: there is no x . X is open
because any ball is by definition a subset of X.
(O2) Let Si be an open set for i = 1, . . . , n, and let x ni=1 Si . We must find an  > 0 for
which B(x, ) ni=1 Si . For each i there is an i such that B(x, i ) Si , because each Si is open.
Let  = min{1 , . . . , n }. Then 0 <  5 i for each i; therefore B(x, ) B(x, i ) Si for each i
i.e., B(x, ) ni=1 Si .
(O3) Let x A S ; we must find an  > 0 for which B(x, ) A S . Since x A S ,
we have x S for some
. Since S is open, there is an  > 0 such that B(x, ) S A S .
k
Definition: A subset S of a metric space (X, d) is closed if it is the complement of an open set.
Theorem:
(C1) and X are closed sets.
(C2) If S1 , S2 , . . . , Sn are closed sets, then ni=1 Si is a closed set.
(C3) Let A be an arbitrary set. If S is a closed set for each A, then A S is a closed set.
In other words, the intersection of any collection of closed sets is closed.
Proof:
(C1) follows directly from (O1).
(C2) and (C3) follow from (O2) and (O3) by De Morgans Laws. k
Exercise: Use De Morgans Laws to establish (C2) and (C3).

Definition: A limit point of a set S in a metric space (X, d) is an element x X for which
there is a sequence in S that converges to x.
In the following theorem, limit points provide an important characterization of closed sets. As
always with characterizations, this characterization is an alternative definition of a closed set.
In fact, many people actually use this as the definition of a closed set, and then the definition
were using, given above, becomes a theorem that provides a characterization of closed sets as
complements of open sets.
Theorem: A set is closed if and only if it contains all its limit points.
Proof:
(): Let S be a closed set, and let {xn } be a sequence in S (i.e., n N : xn S) that converges
to x X. We must show that x S. Suppose not i.e., x S c . Since S c is open, there is an
 > 0 for which B(
x, ) S c . Since {xn } x, let n
be such that n > n
d(xn , x) <  > 0. Then
for n > n
we have both xn S and xn B(
x, ) S c , a contradiction.
(): Suppose S is not closed. We must show that S does not contain all its limit points. Since
S is not closed, S c is not open. Therefore there is at least one element x of S c such that every ball
B(
x, ) contains at least one element of (S c )c = S. For every n N, let xn B(
x, n1 ) S. Then
we have a sequence {xn } in S which converges to x 6 S i.e., x is a limit point of S but is not
in S, so S does not contain all its limit points. k
Example 1: For each n N, let Sn be the open set ( n1 , n1 ) R. Then
n=1 Sn = {0}, which
is not open. This is a counterexample which shows that (O2) would not necessarily hold if the
collection werent finite.
Example 2: For each n N, let Sn be the closed set [ n1 , n1
] R. Then
n=1 Sn = (0, 1), which
n
is not closed. This is a counterexample which shows that (C2) would not necessarily hold if the
collection werent finite.
Example 3: Rn++ is open. This can be shown directly, by finding an appropriate  > 0 for each
x R. Alternatively, one could show that for each i = 1, . . . , n the set Si = {x Rn | xi > 0} is
open, and then invoke (O2) for the set Rn++ = ni=1 Si .
Example 4: The union of all open subsets of Rn+ is an open set, according to (O3). Note that
this set is Rn++ . This is therefore a third way to show that Rn++ is an open set.

Exercise: Is `
++ an open subset of ` ? Prove that your answer is correct.

 be the union of all

Example 5: Generalizing Example 4, let G be any subset of (X, d) and let G
is an open set. G
is clearly the largest open subset of
open subsets of G. According to (O3), G
is itself an open subset of G, and (ii) every open subset of G is a subset
G, in the sense that (i) G
i.e.. G
contains every open subset of G, which we could state informally as G
is at least
of G
the interior of G, also denoted int G.
as large as every other open subset of G. We call the set G
Example 6: Doing the same thing for closed sets, let G be any subset of (X, d) and let G be
the intersection of all closed sets that contain G. According to (C3), G is a closed set. It is the
smallest closed set containing G as a subset, in the sense that (i) G is itself a closed set containing
G, and (ii) every closed set containing G as a subset also contains G as a subset every other
closed set containing G is at least as large as G. We call G the closure of G, also denoted cl G.
The following definition summarizes Examples 5 and 6:
is the union of
Definition: Let G be a subset of (X, d). The interior of G, denoted int G or G,
all open subsets of G, and the closure of G, denoted cl G or G, is the intersection of all closed
sets that contain G.
G G i.e., int G G cl G.
Remark: G
The interior of the complement of G i.e., int Gc is called the exterior of G:
Definition: Let G be a subset of (X, d). The exterior of G, denoted ext G, is the interior of Gc .
Remark: The exterior of G is the union of all open sets that do not intersect G i.e., the largest
open set in Gc .
The boundary of a set lies between its interior and exterior:
Definition: Let G be a subset of (X, d). The boundary of G, denoted bdy G, is the complement
of int G ext G i.e., bdy G = [int G ext G]c .
Remark: The interior, exterior, and boundary of a set comprise a partition of the set. The interior
and exterior are both open, and the boundary is closed. bdy G = cl G cl Gc .

Example 7: Let u : R2++ R be defined by u(x1 , x2 ) = x1 x2 , and let S = {x R2++ | u(x) < }
for some R++ . Note that S = u1 ((, )), the inverse image under u of the open interval
(, ) in fact, S is the inverse image of the open interval (c, ) for any c 5 0. We will show
that S is an open set.
Proof: Let x S; we must find an  > 0 such that B(
x, ) S i.e.,. such that
x B(
x, ) x1 x2 < .

(2)

Lets make the arithmetic easier here by exploiting the fact that kzk 5 kzk2 for all z R2 , so
that B2 (
x, ) B (
x, ) for all x R2++ and all  > 0 i.e.,
x B2 (
x, ) x B (
x, ).
Therefore, if we can find an  > 0 such that
x B (
x, ) x1 x2 < ,

(3)

that will be sufficient to establish (2).

We can further simplify by noting that
x B (
x, ) [x1 < x1 +  & x2 < x2 + ],
so now if we can find an  > 0 such that
[x1 < x1 +  & x2 < x2 + ] x1 x2 <

(4)

then that will be sufficient to establish (3), and therefore (2). Since all x1 and x2 are positive, we
always have x1 x2 < (
x1 + )(
x2 + ), so if we can find an  > 0 such that
(
x1 + )(
x2 + ) <

(5)

then that will be sufficient to establish (4), and therefore, in turn, (3) and (2).
In order to solve the inequality (5), we solve for  in the equation
(
x1 + )(
x2 + ) = .

(6)

2 + (
x1 + x2 ) + (
x1 x2 ) = 0

(7)

Rewriting the equation as

and applying the quadratic formula, we obtain
i
p
1h
2
=
(
x1 + x2 ) (
x1 + x2 ) + 4 4
x1 x2 .
(8)
2
Since x1 x2 < , the positive root yields a positive value of  call this b
 a value that satisfies
(6). Therefore any smaller value of  will satisfy (5), and will therefore in turn satisfy (4), and
therefore (3), and therefore (2). k
4

Example 7 is an important example not the specific utility function u(x1 , x2 ) = x1 x2 , but the
fact that the set S we defined (the inverse image under u of an open set in R) was itself an open
set in the domain of u. In fact, its a theorem well do later that a function f : (X, d) (Y, d 0 )
is continuous if and only if f 1 (A) is open in X for every open set A Y . This has important
implications for utility theory and other applications in economics.
But suppose we change the example so that u : R2+ R i.e., weve changed the domain of u
from R2++ to R2+ . Now the set u1 ((, )) includes the boundary of the set R2+ , i.e., the axes of
R2+ . The inverse image of the set (, ) does not appear to be an open set, as it was before. But
the function u(x1 , x2 ) = x1 x2 doesnt seem any less continuous than it was before and indeed,
when we define continuity, as were going to do shortly, well find that u is continuous, even with
the domain R2+ .
The key to this puzzle is in the condition, two paragraphs above, that for every open set A in the
target space Y , the set f 1 (A) must be open in X i.e., open in the domain of f . In Example
7, the domain of u is R2++ ; in our variation on the example, the domain of u is R2+ . Each of these
sets is a metric space, with the metric it inherits from R2 , but they are not the same metric space
as R2 . Therefore a set might be open in one of these three metric spaces but not in another.
For example, let x = (0, 2) R2 and let r = 1, and consider how the open ball B(x, r) changes
as we move from the metric space R2 to the metric space R2+ to the metric space R2++ . Recall the
definition of the open ball B(x, r):
Definition: Let (X, d) be a metric space, let x X, and let r be a positive real number. The
open ball about x of radius r is the set B(x, r) := {x X | d(x, x) < r}.
As the metric space playing the role of the definitions (X, d) changes from X = R2 to X = R2+ to
X = R2++ , the set B(x, r) changes from, first, the whole disc about the point (0, 2) of radius r = 1,
to the half-disc which is the intersection of the whole disc and R2+ , to a set thats not defined
/ X. But undefined sets like this did not appear
the set is not defined in the third case because x
in our proof in Example 7, because all the -balls we used were centered at points x R2++ .
In order to take a general approach to this issue, lets suppose we begin with a metric space (X, d),
but then we want to work in a subspace (X 0 , d 0 ), where X 0 is a subset of X and d 0 is the metric
that X 0 inherits from (X, d). For any open ball B(x, r) in X, let B 0 (x, r) be the corresponding
open ball in X 0 i.e., B 0 (x, r) := B(x, r) X 0 .

In our example we started with the metric space X = R2 and then we moved to the domain of u,
which was the subspace X 0 = R2++ X. Everything worked OK here because the set X 0 was an
open subset of X, so for any point x X 0 , there is an open ball B(x, r) about x that lies entirely
in X 0 therefore B 0 (x, r) = B(x, r). Any subset S X 0 is open in the metric space X 0 if and
only if its open in the metric space X.
But when we moved instead to the subspace X 0 = R2+ X, that set was not an open set in
the original metric space X = R2 . Therefore there were sets, like the half-disc B 0 (x, r) i.e.,
B(x, r) R2+ that are open in the metric space X 0 but not open in the original metric space X.
This is a common occurrence. A useful concept to clarify it is the idea that a subset S of a subspace
X 0 of X can be open relative to X 0 :
Definition: Let (X, d) be a metric space, let X 0 X, and let S X 0 . The set S is open relative
to X 0 if S is an open set in the metric space X 0 , with the metric it inherits from (X, d). We also
say simply that S is open in X 0 .
The following remark provides a useful characterization of relatively open sets.
Remark: If S is a subset of X 0 X, then S is open relative to X 0 if and only if there is an open
set V in X such that S = V X 0 .
Proof:
(): Suppose S is open relative to X 0 . Then for every x S there is an x > 0 such that
B 0 (x, x ) S. But also B(x, x ) is open in X. Let V = xS B(x, x ); then V is open in X and
S = V X 0 , as follows:
xS B 0 (x, x ) = S
i.e., xS [B(x, x ) X 0 ] = S
i.e., [xS B(x, x )] X 0 = S
i.e., V X 0 = S.
(): Suppose V is an open set in X such that S = V X 0 . We must show that S is open
relative to X 0 . Let x S and let  > 0 be such that B(x, ) V [which we can do because V is
open in X]. Then also B 0 (x, ) V , and of course B 0 (x, ) X 0 . Therefore B 0 (x, ) V X 0
i.e., B 0 (x, ) S. Since we chose x S arbitrarily, this establishes that S is open relative to X 0 .
k
Of course, we can do exactly the same thing for closed sets:
6

Definition: Let (X, d) be a metric space, let X 0 X, and let S X 0 . The set S is closed
relative to X 0 if S is a closed set in the metric space X 0 , with the metric it inherits from (X, d).
We also say simply that S is closed in X 0 .
And we have the parallel characterization of relatively closed sets:
Remark: If S is a subset of X 0 X, then S is closed relative to X 0 if and only if there is a closed
set G in X such that S = G X 0 .
Proof: Exercise.