You are on page 1of 22

General Topology

Summer, 2011
1 Introduction
These notes present some theorems and concepts from General Topology.
We begin with the denitions of topology, base, continuity, and lterbases. Then we
examine a few properties of compactness. Next, we examine product spaces and prove
Tychonos theorem. We consider a handful of separation axioms such as normality and
give Urysohns characterization of normality.
We also consider the notion of a connected space and show that for compact hausdor
spaces, totally disconnected and having a clopen base are equivalent. Compact totally dis-
connected hausdor spaces are examined further by looking at Stone duality.
Finally, we consider the Baire categeory theorem and the Kuratowski-Ulam theorem.
The texts from which this material came include: Topology by Dugundji, Measure and
Category by Oxtoby, Real Analysis by Royden, and Metric Spaces by Wieting.
2 Basic Denitions
Topology A topology is on a set X is a collection T P(X) of subsets of X such that
1. T is closed under arbitrary unions
2. T is closed under nite intersections
3. , X T
The elements of T are called the open sets. A set is closed if its complement is open. One
can also dene a topology in terms of the closed sets. Then T would be closed under nite
unions and arbitrary intersections but otherwise the denition is the same. A set which is
both closed and open is called clopen. A set which is neither closed nor open has no special
name that Im familiar with; I suppose we can call it nooclo.
Generating a Topology Let A P(X). Then P(X) is a topology containing A, and
the intersection of topologies containing A is itself a topology containing A, the smallest
topology containing A, and well denote it T
A
. T
A
can also be described by saying it consists
of unions of nite intersections of elements of A, plus and X.
Bases A set B T is called a base for T if every open set is a union of elements of B.
The elements of B are called basic open sets. Of course bases are not necessarily unique.
If A has the property that all nite intersections of elements of A are also unions of
elements of A (e.g., A is closed under nite intersections), then A , X forms a base for
T
A
. Of course if , X A, then A itself forms a base for T
A
.
1
If A has the property that T
A
= T, i.e. A generates T, then we say that A is a subbase
for T.
An Example Our quintessential example of a topological space is the following. Let L be
a logic. I.e., for each signature , L() is a class of -structures (called L()-sentences). We
typically assume L is closed under boolean combinations. (See the section on Lindstr oms
theorem in the Model Theory notes for more discussion, or a similar section in the Modal
Logic notes.) We dene a topology on the class of all -structures as follows. We declare
that the basic open sets are the L()-sentences. Under the assumption that L is closed
under boolean combinations, we have = as an L()-sentence, and similarly = X is
an L()-sentence. Further, L()-sentences are closed under intersection, so we they form a
base for the topology they generate. In fact, this is a clopen base, because of the assumption
of closure under .
Compactness A topological space is said to be compact if for every open covering of X
there is a nite subcovering. This is equivalent to saying that every basic open covering of
X has a nite subcovering. It is also equivalent to saying that every collection of closed sets
with the nite intersection property (i.e., nite intersections are nonempty) has nonempty
intersection. Again, this is equivalent to the same statement but said of basic closed sets.
When said of a logic, we typically express this condition by saying that every collection of
sentences (i.e. basic closed sets) which is nitely satisable is satisable.
Continuous Functions The morphisms of topological spaces are continuous functions. A
function f : X Y is continuous means that for every open U Y , f
1
(U) X is open
too. Another way to say this is that f is continuous at every point x X: this means that
for every open V Y containing f(x), there is some open U X with x U and f(U) V .
The intuition here is that if x

is close to x, then f(x

) is close to f(x).
If A is a subbase, then to show that f is continuous it suces to show that for every
a A, f
1
(a) is open. To see this, note that of course f
1
(Y ) = X and f
1
() = are
open. For other open U Y , we may introduce a collection J of subsets of Y such that
U =

jJ
j and each j is a nite intersection of elements of A. Thus, we may introduce, for
each j J, elements a
j1
, . . . a
jn
j
of A such that j =

n
j
k=1
a
jk
. I.e.
U =
_
iJ
n
j

k=1
a
jk
It follows that
f
1
(U) =
_
iJ
n
j

k=1
f
1
(a
jk
)
and so if f
1
(a
jk
) is always open, f
1
(U) is open too.
2
Nets We can also express this condition using nets or lterbases. A net is a function from
a directed set to X. A directed set is a set D together with a preorder (transitive and
reexive) that satises the following property: for every a, b D, there is an element c D
such that a, b c. An example of a directed set is (N, ) with the usual ordering, or the
collection of all open sets containing a point x X ordered under reverse-inclusion. A net
is a function from a directed set to X. A net f : D X is said to converge to a point y in
X if for every open set U containing y there is some a D such that for all b a we have
f(b) U.
A net does not necessarily converge to a unique limit, even if it converges. In fact, every
net having at most one limit is equivalent to the space being hausdor. X is hausdor means
that for every x, y X with x ,= y there are open sets U and V such that x U, y V ,
and U V = . I.e. we can separate distinct points with disjoint open sets. Suppose that X
is hausdor and f : D Y converges to y
1
and y
2
. If y
1
,= y
2
, we could introduce disjoint
open sets separating them, and this would be contradictory. Now suppose that every net has
at most one limit, and let x
1
, x
2
X be elements which cannot be separated with disjoint
open sets. Let
D = U
1
U
2
[ x
1
U
1
and x
2
U
2
and U
1
, U
2
open
and order D under reverse-inclusion. Then D is a directed set. Since each U
1
U
2
is
nonempty, we may select a point f(U
1
U
2
) from each. f is a net which converges to both
x
1
and x
2
, and so x
1
= x
2
.
The Net Version of Continuity A function f : X Y between topological spaces is
continuous at x i for every net x
d
(d D) which converges to x, the net f(x
d
) (d D)
converges to f(x). To see this, let f be continuous at x and let x
d
converge to x. Let
V Y be an open set containing f(x). Since f is continuous at x, we may introduce an
open set U X such that x U and f(U) V . Since x
d
converges to x, we may introduce
an element c D such that for all d c we have x
d
U. Then, for all d c, we have
f(x
d
) f(U) V . Thus, f(x
d
) converges to f(x).
Now suppose that every net x
d
converging to x has f(x
d
) converging to f(x). Let V
be an open set containing f(x). Suppose to get a contradiction that for every open set U
containing x, there is some element x
U
such that f(x
U
) , V . Let D be the collection of all
open sets containing x ordered under inclusion. Then x
U
(U D) is a net converging to x.
As such f(x
U
) converges to f(x). Yet f(x
U
) is not eventually in V , a contradiction.
Sequences and Metric Spaces A sequence is a special kind of net where the directed set
D is N ordered as usual. For metric spaces its sucient to focus on sequences in the sense
that the above characterization of continuity still holds. A metric space is a set X together
with a binary function d: X
2
R such that
1. d(x, y) 0
2. d(x, y) = 0 x = y
3
3. d(x, y) = d(y, x)
4. d(x, y) +d(y, z) d(x, z)
Thus d is an abstract notion of distance. Every metric space induces a topology. For any
x X and r R, we set B(x, r) := y X [ d(x, y) < r. Let B = B(x, r) [ x X, r R.
Then B generates some topology T = T
B
. In fact B is a basis for T. To see this, we note that
B(x, 0) = , so B. Further, X =

rR
B(x, r), so X is the union of elements of B. It
remains to show that any nite intersection of elements of B is a union of elements of B. For
this it suces to show that any element in an intersection of two elements of B is contained
in some element of B which is a subset of the intersection. So let x B(y
1
, r
1
) B(y
2
, r
2
).
Let r = min(r
1
d(x, y
1
), r
2
d(x, y
2
)). This ensures x B(x, r) B(y
1
, r
1
) B(y
2
, r
2
).
Now lets see that sequences are enough. I.e., we show that f : X Y (where X and
Y are two metric spaces) is continuous at x X i for every sequence x
n
that converges to
x, f(x
n
) converges to f(x). The left-to-right direction follows from our previous observation
for nets. So suppose that for every sequence x
n
converging to x we have f(x
n
) converging
to f(x). Let V be an open set containing f(x). We wish to nd an open set U containing x
such that f(U) V . We note that x
n
converges to x i x
n
is eventually in any basic open
set containing x, and this in turn is equivalent to x
n
eventually being in any nonempty ball
centered at x (as B(x, r d(y, x)) B(y, r)). So, suppose to get a contradiction that for
each n N there is some element x
n
B(x,
1
n
) such that f(x
n
) , V . Then x
n
converges
to x, and so by assumption f(x
n
) converges to f(x). Thus, f(x
n
) is eventually in V , a
contradiction.
In fact, the argument shows that sequences are enough whenever, for any point x X,
there is a countable collection O
i
(i N) of open sets containing x such that for any open
set U containing x, there is some O
i
U. This condition is called rst countable. I.e. X is
rst countable means that for every x X, there is a countable neighborhood base for x
in the above sense.
Homeomorphism A bijective continuous function with continuous inverse is called a
homeomorphism, and this is the topological notion of isomorphism because if X and Y
are related by some homeomorphism f, then f is also simultaneously a bijection on the
topologies. I.e. f : T(X) T(Y ) dened by O f(O) is a bijection.
Closure and Interior The closure

Y of a subset Y of a topological space X is the smallest
closed set containing Y . The intersection of all the closed sets containing Y is itself a closed
set containing Y , and so this is a well-dened notion.
The interior int Y of a subset Y of a space X is the largest open subset of Y . The union
of all the open sets conatined in Y is itself an open set contained in Y , so this is well-dened.
The boundary of Y is the closure minus the interior, and the exterior is the complement
of the closure.
One can check the following properties:
1.

Y is the set of all points x such that any open set containing x intersects Y .
4
2. int Y is the set of all points x for which there is an open set O with x O and O Y .
3. The interior of Y is the exterior of Y .
4. The boundary of Y is same as the boundary of Y .
5. The interior and the boundary and the exterior for any Y partition X.
Subspaces Given any subset Y of a space X, we may put a topology on Y : O Y [
O open in X. This is called the subspace topology. One thing to observe about this def-
inition is that if f : Z X is continuous, and Y X is the range of f, then f is also
a continuous function f : Z Y where Y is given the subspace topology. Similarly, if
f : Z Y is continuous, then f : Z X is continuous.
Filterbases Filterbases can play the same role that nets play in generalizing the notion
of convergence from metric spaces to topological spaces. A lter base B on a space X is a
collection of subsets of X such that
1. For every b
1
, b
2
B there is some b
3
B such that b
3
b
1
b
2
2. , B
Given any net f : D X, we may form a lterbase as follows. For each d D, we may
dene D
d
:= c D [ c d, and we may dene B
f
:= f(D
d
) [ d D. Now, B
f
is
a lterbase: f(D
d
) is not empty because, e.g., f(d) f(D
d
), and given f(D
d
1
) and f(D
d
2
)
we may obtain a d
3
d
1
, d
2
, and as is transitive, we get that f(D
d
3
) f(D
d
1
) f(D
d
2
).
Note that f converges to x i for every open set O containing x there is some b B
f
such
that b O. This motivates the following denition of convergence for lterbases: B is said
to converge to x if for every open set O containing x there is some b B such that b O.
Now, we may also form a net from any lterbase with the same convergence properties.
Given a lterbase B, set D = (x, b) [ b B, x b and set (x, b) (y, c) i c b. This is
clearly transitive, and directedness is exactly the assumption that given b
1
, b
2
there is a b
3
such that b
3
b
1
b
2
. We dene a net f
B
: D X by setting f(x, b) = x. We claim that f
B
converges to x i B converges to x. First, assuming that B converges to x, we try to show
that f
B
converges to x, i.e. for every open set O containing x there is some d D such that
for all d

d, f(d

) O. Well, given such an O, and since , B, we may introduce such a b


and an x b. Then, for any (y, c) (x, b), we have that c b O and so y O, i.e. f
B
is
eventually in O. Conversely, given that f
B
converges to x, and given an open O containing
x, we get (x, b) such that, in particular, for all y b, f
B
(y, b) O. Thus, b O, and so B
converges to x.
We have described two maps, one takes a net f and forms a lterbase B
f
, and the
other takes a lterbase B and forms a net f
B
. We have seen that both operations preserve
convergence. It follows that every lterbase converges to at most one limit is equivalent to
being hausdor, by our previous result for nets.
5
Also, B
f
C
= C for any lterbase C. I.e., B

is a left inverse of f

, and so f

is injective.
To see this, suppose b C. Then let x b, and note that f
C
((y, c) [ (y, c) (x, b)) = b.
Thus, b B
f
C
. Conversely, given that b B
f
C
we may introduce (y, c) such that b = y

[
c

C[(y

, c

) (y, c)]. But then b = c and so b C. So, we see that nets can be thought
of as a generalization or expansion of lterbases; yet, then again, lterbases abstract away
properties of the directed set which may not be relevant.
Its not the case that f
Bg
= g for every net g. For example, take a constant sequence
g : N X where g(n) = x for all n N. Then B
g
= x, and so the domain of f
Bg
is
(x, x).
In what follows, we shall be formulating things in terms of lterbases rather than nets,
but similar results would typically hold for them as well.
Given any mapping f : X Y and a lterbase B on X, we may form a lterbase f(B)
on Y by setting
f(B) := f(b) [ b B
This is a lterbase. Further, f is continuous at x i every lterbase B P(X) converging to
x also has f(B) converging to f(x). To see this, assume rst that f is continuous at x and B
converges to x. Let V be an open set containing f(x). Then let U be an open set containing
x such that f(U) V . Then introduce a b B such that b U. We have f(b) f(U) V
and so f(B) converges to f(x).
Conversely, assume that every B converging to x has f(B) converging to f(x). Then
consider an open set V containing f(x). Since the collection B of open sets containing x
forms a lterbase converging to x, we know by assumption that f(B) converges to f(x).
Thus, we may introduce a U B such that f(U) V .
A lterbase B is said to be a sublterbase of a lterbase C if for every c C there is some
b B such that b c. In particular, if B C, then B is a sublterbase of C. Note that, by
the ultralter lemma, any lterbase may be extended to an ultralter, and so every lterbase
has a sublterbase which is an ultralter. Also note that a lterbase B converges to x i B
is a sublterbase of the collection of all open sets containing x (which is a lterbase).
A lterbase B is said to cluster at x, or that x is a cluster point of B, if for every open
set O containing x, and for every b B, we have Ob ,= . Every limit of B is also a cluster
point of B. However, not every cluster point is a limit. E.g., set B = x, y where x ,= y
are two distinct points in some hausdor space. Then B clusters at both x and y, but it
converges to neither.
If B is a sublterbase of C and C converges to x, then so too does B. Similarly (yet in
the other direction), if B clusters at x, then so too does C.
Another way to say that B clusters at x is to say that x

b [ b B. An ultralter
converges to x i the ultralter clusters at x. To see this, let U cluster at x and let O be an
open set containing x. Then O must be in U lest X O is in U and x isnt a cluster point.
Let Y be a subset of a space X. We note that x

Y i there is a lterbase B such that
B P(Y ) and B converges to x. First assume x

Y . I.e., every open set containing x
intersects Y . Let B consist of O Y where O is any open set containing x. Then B is a
lterbase and B is contained in P(Y ). Further, given any open set O containing x, OY is
6
an element of B contained in O, so B converges to x. Conversely, let B P(Y ) such that
B converges to x. Let O be an open set containing x, and we show that O Y ,= . Let
b B such that b O. Then b Y and b ,= and so O Y ,= .
Quotient Spaces Given f : X Y where X is a topological space, Y is a set, and f is
any map, then we can form the largest topology on Y such that f is continuous. We declare
U Y open i f
1
(U) is open. If this denes a topology, of course it is the largest possible
topology on Y such that f is continuous. To see that it does dene a topology, we note
f
1
(Y ) = X, and f
1
() = , so Y and are indeed open. Further, suppose f
1
(U
i
) is open
for each i I. Then f
1
(

iI
U
i
) =

iI
f
1
(U
i
) is the union of open sets and therefore
open. Similarly for nite intersections.
Given an equivalence relation E on X, we get a surjection p: X X/E which sends
each element x of X to its equivalence class p(x). The quotient topology is dened as the
largest topology on X/E such that this surjection is continuous. and so we get the quotient
topology as above.
The quotient topology has the following universal property. If f : X Z is any contin-
uous function that respects the equivalence relation (i.e. xEy implies f(x) = f(y)), then
there is a unique continuous function g : X/E Z such that gp = f. To see this, we dene
g(p(x)) = f(x). This is well-dened because f respects E. Of course g will be unique. It
remains to show that g is continuous. Let V be an open subset of Z. Then g
1
(V ) is open
i, by denition, p
1
(g
1
(V )) is open. But this latter set is simply (g p)
1
(Y ) = f
1
(V ),
which is open since f is assumed to be continuous.
3 Compactness
Proposition 1. The following conditions on a space X are equivalent:
1. X is compact
2. Every lterbase of X has at least one cluster point
3. Every ultralter of X converges
Proof. Suppose X is compact, and let B be a lterbase. We need to show that

bB

b
contains an element. Well, every nite collection

b
1
, . . . ,

b
n
is satisable because b
1
, . . . , b
n
is
satisable.
Now suppose that every lterbase has at least one cluster point, and let U be an ultralter.
Then U, being a lterbase, has a cluster point, and, being an ultralter, converges to that
point as well.
Now suppose that every ultralter converges, and we show that X is compact. Let C
i
(i I) be a consistent collection of closed sets (i.e., no nite intersection yields ). Then
C
i
may be extended to an ultralter U, by the ultralter lemma. Let U converge to x. In
particular, U clusters at x, so we know x

b [ b U. Since C
i
= C
i
, we get that x C
i
for each i.
7
Proposition 2. 1. The continuous image of a compact space is compact.
2. A closed subset of a compact space is compact.
3. A compact subspace of a compact, hausdor space is closed.
Proof. Let f : X Y be continuous, and let X be compact. We might as well assume that
f is surjective, because f is still continuous when we shrink the codomain to the size of the
image. Let U
i
(i I) be an open cover of Y . Then f
1
(U
i
) (i I) is also an open cover
of X. Let f
1
(U
1
), . . . , f
1
(U
n
) be a nite subcover (of X). It follows that U
1
, . . . , U
n
is a
cover of Y because f is a surjection.
Let Y be a closed subset of a compact space X. Let O
i
Y be an open cover of Y
(where the O
i
are open sets of X). Since Y is closed, X Y is open, and X Y , O
i
(i I)
is an open cover of X. Let X Y, O
1
, . . . , O
n
be a nite subcover (of X). It follows that
O
1
Y, . . . , O
n
Y is a cover of Y .
Now let Y be a compact subspace of a compact, hausdor space X. To show that XY
is open, it suces to show that for each x X Y there is some open set U such that
x U and U Y = . Let x X Y . For each y Y , as X is hausdor, we may introduce
open sets O
y
and U
y
such that y O
y
, x U
y
and O
y
U
y
= . Since

yY
O
y
Y , and
Y is compact, we may introduce y
1
, . . . , y
n
in Y such that O
y
1
O
yn
Y . Then let
U := U
y
1
U
yn
. Then U is as desired.
4 Tychonos Theorem
Product Spaces Let X
i
(i I) be a collection of topological spaces. We may form the
set

iI
X
i
as normal, and have the usual projections p
i
:

iI
X
i
X
i
. We shall now
dene a topology on this cartesian product, called the product topology, with the following
properties. Each p
i
will be continuous, and for any topological space Y with continuous
functions f
i
: Y X
i
, there exists a unique continuous function f : Y

iI
X
i
such that
p
i
f = f
i
for each i I. I.e.,

iI
X
i
together with the projections is the categorical
product in the category of topological spaces and continuous functions.
We declare that
A := p
1
i
(U) [ i I, U X
i
, U open
is a subbase for the product topology on X :=

iI
X
i
. Since A already contains X and ,
we note that a subset of X is open i it is the union of nite intersections of elements of A.
By construction, we have that each p
i
is continuous. So consider a space Y with continuous
functions f
i
: Y X
i
. We dene f : Y X be setting f(y) = where (i) = f
i
(y). Of
course we get that p
i
f = f
i
for each i. Also, f is unique in this property.
What we need to check is that f is continuous. As A is a subbase, it suces to show
that f
1
(a) is open for each a A. Well, write a = p
1
i
(O) for some i I and some open
set O X
i
. Well, f
1
(p
1
i
(O)) = (p
i
f)
1
(O) = f
1
i
(O).
8
If I
0
I and O
i
is a subset of X
i
for each i I
0
, then we use

iI
0
O
i
to denote the set
of X such that (i) O
i
for each i I
0
. We claim that
B := O
i
1
O
in
[

i is a nite tuple from I and O


i
j
X
i
j
is open
is base for X. This is clear because B is just the collection of nite intersections of elements
of A, i.e.
O
i
1
O
in
=
n

j=1
p
1
i
j
(O
i
j
)
Now, if I is nite, say I = 2, and X = X
1
X
2
, then a base for the product topology is
given by
B = O
1
O
2
[ O
1
X
1
, O
2
X
2
both open
This is the same base as above. It should be remarked, however, that this base does not
work in the case I is innite. I.e., if I is innite,
B

:=

iI
O
i
[ O
i
X
i
is open for each i I
is not a necessarily a base for X. In fact, if U
i
(i I) is a collection of nonempty open sets
of X
i
(respectively), and if i I [ U
i
,= X
i
is innite, then

iI
U
i
is not open. As

iI
U
i
is nonempty, it suces to show that there is no nonempty basic open set O
i
1
O
in
which is a subset of

iI
U
i
. Given a nonempty basic open set O
i
1
O
in
, we observe
that there are innitely many and so at least one i I such that i ,= i
j
for j = 1, . . . n and
U
i
,= X
i
. Thus, we may select some x X
i
U
i
and form a O
i
1
O
in
such that
(i) = x. Then ,

iI
U
i
as desired.
Another fact to be noted is that each projection map is open in the sense that it sends
open sets to open sets.
Lemma 3. Let B be a lterbase on a product space X =

iI
X
i
. Then B converges to
X i p
i
(B) converges to (i) for each i I.
Proof. Suppose that B converges to . Then, since p
i
is continuous, we have that p
i
(B)
converges to p
i
() = (i).
Conversely, suppose that p
i
(B) converges to (i) for each i. It is enough to show that B
is eventually in any basic open set containing . So, let O
i
1
O
in
be a basic open set
of X containing . Then we know that (i
j
) O
i
j
for each j = 1, . . . , n. Thus, as p
i
j
(B)
converges to (i
j
) we may introduce b
1
, . . . , b
n
B such that p
i
j
(b
j
) O
i
j
. Then it follows
that
b
1
b
n
O
i
1
O
in
Finally, we may introduce a b B such that b b
1
b
n
, since B is a lterbase.
Theorem 4 (Tychono). The product space X :=

iI
X
i
is compact i X
i
is compact for
each i I.
9
Proof. Suppose that X is compact. Then, as each projection p
i
is a continuous surjection,
we get from Proposition 2 that each X
i
is compact.
Now suppose that each X
i
is compact. We would like to show that every ultralter of
X converges; this is equivalent to compactness by Proposition 1. So let U be an ultralter
on X. Since each projection p
i
is surjective, it follows that p
i
(U) = p
i
(b) [ b U is an
ultralter on X
i
. As each X
i
is compact, we may introduce x
i
X
i
such that p
i
(U) converges
to x
i
. The we dene X by (i) = x
i
. By Lemma 3, we obtain that U converges to .
Compactness of Propositional Logic Tychonos theorem about the compactness of
product spaces may be used to prove the compactness theorem for propositional logic. First
lets recall what propositional logic is. We have a set = p
i
[ i I of proposition letters.
We have L = , , an algebraic signature. And the set of formulas F is the free L-
algebra with as generators. A valuation is a map v : 2, and as F is the free L-algebra
with as generators, valuations extend uniquely to L-homomorphisms v from F to 2.
Let X be the set which consists of all the valuations, i.e. maps v : 2. For each
F we let B

:= v [ v() = 1. Since B

2
= B

1
B

2
and B

= and B

= X, we
see that B = B

[ F forms a base for a topology T on X. Further B is a clopen base,


since X B

= B

.
T is compact in the topological sense i F is compact in the logical sense. In detail, T is
compact i every collection of basic closed sets that is consistent has nonempty intersection.
Since B is a closed base, this is equivalent to saying that for every F, the existence of a
valuation v for each nite subset
0
of that satises
0
implies the existence of a valuation
v that satises all of . This is the logical version of compactness for F.
Let the L-algebra 2 be given the discrete topology (all sets are open). Of course, being
nite, 2 is compact. And so 2

v
2 is compact by Tychonos theorem. However, we
claim that 2

is homeomorphic to X, from which it follows that X is compact as well. First


note that the underlying sets of 2

and X are the same, and so it suces to show that the


identity map id: 2

X is continuous, and its inverse is continuous.


To see that id
1
is continuous, we give ourselves a basic open set O
i
1
O
in
of 2

and
show that it is open in X. We may assume that each O
i
j
is either 0 or 1 (2 = 0, 1),
because if O
i
j
= 2, we can remove it without changing the basic open set (or its the only
thing to remove and the basic open set is 2

= X and so open in X), and if its , then the


basic open set is and so open in X by default. Now dene J 1, . . . , n such that j J
i O
i
j
= 1. Then dene an element of F by:
:=

jJ
p
i
j

j{1,...,n}J
p
i
j
It follows that B

= O
i
1
O
in
.
Conversely, to see that id is continuous, its enough to show that the inverse image of
each basic open set in X is open in 2

. I.e., we want to show that B

is open in 2

for each
10
F. We may write in disjunctive normal form

j
p
ij
and then for each p
ij
with a + there is a corresponding open set O
ij
= p
1
p
ij
(1) and
similarly for each p
ij
with a there is O
ij
= p
1
p
ij
(0). Then
B

=
_
i

j
O
ij
is open in 2

.
5 Separation Axioms
A space is hausdor if every pair of distinct points x
1
and x
2
can be separated by disjoint
open sets. I.e., there exists disjoint open sets O
1
and O
2
such that x
1
O
1
and x
2
O
2
. A
space is regular if points can be separated from closed sets by open sets. I.e., if C is a closed
set and x , C, then there exists disjoint open sets O
1
and O
2
such that x O
1
and C O
2
.
Finally, a space is said to be normal if closed sets can be separated from closed sets by open
sets. I.e., if C
1
and C
2
are disjoint closed sets, then there are disjoint open sets O
1
and O
2
such that C
1
O
1
and C
2
O
2
.
Every hausdor normal space is regular, because in a hausdor space singletons are closed
sets.
Proposition 5. Every compact hausdor space Y is normal.
Proof. First we show that Y is regular. Let C be a closed set and y , C. For each x C
let (U
x
, V
x
) be disjoint open sets such that x U
x
and y V
x
. As C is a closed subset of a
compact space, it is compact, so we may introduce x
1
, . . . , x
n
such that U
x
1
U
xn
C.
Then the open sets

n
i=1
U
x
i
,

n
i=1
V
x
i
disjointly separate C and y.
Now let C and D be two disjoint closed sets. For each y C, by regularity we may
separate it from D via disjoint open sets (U
y
, V
y
). Then using compactness again we get U
y
1

U
yn
C, and the disjoint open sets

n
i=1
U
y
i
,

n
i=1
V
y
i
give the desired separation.
Theorem 6 (Urysohn). Let X be any topological space. X is normal i there exists, for
every pair of disjoint closed sets (A, B), a continuous function f : X [0, 1] such that
f(A) = 0 and f(B) = 1
Proof. First assume that there is such a function for every pair of disjoint closed sets (A, B).
Then f
1
([0, 1/2)) and f
1
((1/2, 1]) are disjoint open sets that separate A and B.
Now assume that X is normal. Let two disjoint closed sets A and B be given. In order to
dene f, we shall nd it helpful to rst dene for every rational number an open set U
p
of X.
For p < 0, we set U
p
:= , and for p > 1 we set U
p
:= X. Next, we dene U
1
:= B. Since
11
X is normal, we may introduce disjoint open sets U and V such that A U and B V .
Since V is a closed set containing U, we see that

U V B = U
1
. We let U
0
= U.
Once again, we have that A U
0
,

U
0
U
1
, and U
1
B.
So far we have dened U
p
so that if p < q, then

U
p
U
q
. We wish to continue this for
the rationals in (0, 1). Since the rationals are countable, we may list the rationals in (0, 1)
as (p
1
, p
2
, p
3
, . . .), and dene U
pn
assuming U
pm
has been dened for all m < n. Given p
n
,
we note that among 0, 1, p
1
, p
2
, . . . , p
n1
it must have an immediate predecessor p and an
immediate successor p
+
. By the induction hypothesis, we have that U
p
U
p
+. Thus, U
p
and
U
p
+ are disjoint closed sets, and by normality we may introduce disjoint open sets U and
V such that U U
p
and V U
p
+. Letting U
pn
= U, we see that U
p
U
pn
and U
pn
U
p
+
as desired. Of course, if q p and U
q
has been dened, then U
q
U
pn
too, and similarly for
q p
+
.
So, we get for every rational number p an open set U
p
of X such that p < q implies
U
p
U
q
. Further, = U
p
for all p < 0, X = U
p
for all p > 1, and U
0
A and U
1
B.
For each x X we let P(x) = p Q [ x U
p
. For any x X, P(x) is bounded below
by, say, 1, and so the inmum is well-dened. We set
f(x) := inf P(x)
We claim that f is the desired function. First note that if a A, then f(a) = 0, because
P(a) = [0, ). On the other hand, if b B, then f(b) = 1, since P(b) = (1, ). Also, the
range of f is contained in [0, 1].
Now we show that f is continuous. Note rst however that if x U
p
, then for every
q > p we have x U
q
and so f(x) p. Similarly, if x , U
p
, then f(x) p.
We show that f is continuous at x for every x X. Let V be an open set containing f(x).
We need to nd an open set U of X such that f(U) V . Since the open intervals form a
base for the topology on R, we may introduce an open interval (c, d) such that f(x) (c, d).
Since the rationals are dense in the reals, we may introduce rational numbers p and p
+
such
that
c < p < f(x) < p
+
< d
Since f(x) , p we get that x , U
p
. Since f(x) , p
+
, we get that x U
p
+. Let U = U
p
+ U
p
.
This is an open set containing x. Further, if y U, then y U
p
+ and so f(y) p
+
. Similarly,
y , U
p
implies that f(y) p. Thus, f(U) (c, d), as desired.
One can use Urysohns characterization of normality to prove Tietzes characterization
of normality, which states that in a hausdor space, normality is equivalent to saying that
for every closed set A, every continuous function f : A R has a continuous extension
F : X R.
6 Connectedness
A space X is said to be disconnected if there exists a clopen set other than X or . I.e., there
exists a pair of nonempty open sets that partition X. A space X is said to be connected
12
otherwise. As an example, R is connected, yet (0, 1) (2, 3) with the subspace topology is
not connected.
A subset Y of X is connected if, given the subspace topology, is connected. This amounts
to saying that Y is disconnected i there exists open sets U and V of X such that U and V
both intersect Y , U V Y = , and U V Y .
Components A subset Y of X is said to be a component if Y is maximally connected.
I.e., Y is connected yet Z Y and Z is connected implies that Z = Y . The components of
X always partition X. We verify this as follows.
First, note that if x C
1
C
2
and C
1
and C
2
are connected, then C
1
C
2
is connected. To
see this, assume to get a contradiction that U, V are open sets of X such that UC
1
C
2
,= ,
V (C
1
C
2
) ,= , UV (C
1
C
2
) = , and (UV ) (C
1
C
2
). Without loss of generality,
assume x U and V C
1
,= . Then UC
1
,= , V C
1
,= , UV C
1
= , and UV C
1
,
which violates the assumption that C
1
is connected.
It follows that any point x can be in at most one component. Otherwise, we have an x
in two distinct components C
1
and C
2
, from which it follows that C
1
C
2
is a connected set
strictly larger than one of C
1
or C
2
, a contradiction.
Finally, we note that every point is in some component. Let x X. Let
C
x
:=
_
Y X [ x Y and Y is connected
We claim that C
x
is a component that contains x. Of course x is a connected set that
contains x, so x C
x
. Also, C
x
is maximally connected if its connected by construction.
To see that its connected, suppose, to get a contradiction, that there are open sets U and
V of X such that U C
x
,= , V C
x
,= , U V C
x
= , and U V C
x
. Then
since U C
x
,= , we may introduce a connected set Y
U
which contains x and intersects U.
Similarly, since V C
x
,= , we may introduce a connected set Y
V
which contains x and
intersects V . We know that Y
U
Y
V
is connected, because both summands contain x, yet
U V (Y
U
Y
V
), U (Y
U
Y
V
) ,= , V (Y
U
Y
V
) ,= , and U V (Y
U
Y
V
) = , a
contradiction.
In fact, the components are always closed, since the closure of any connected set is
connected. This boils down to observing that if U is open and U

Y ,= , then U Y ,= .
Totally disconnected A totally disconnected space is one in which the components are
singletons.
Proposition 7. Suppose that X is hausdor and compact. Then X is totally disconnected
i X has a base of clopen sets.
Proof. First suppose that X has a base of clopen sets. Then let Y be a set which consists
of two or more elements. Let y
1
, y
2
Y with y
1
,= y
2
. Then, since X is hausdor, we may
introduce disjoint open sets separating y
1
and y
2
. Thus, there is a clopen set containing y
1
but not y
2
. So, Y isnt connected.
13
Now suppose that X is totally disconnected. Let O be open and let y O. We wish to
nd a clopen set C
y
such that y C
y
O. Then O =

yO
C
y
and so the clopen sets will
form a base. It suces to show that

C = y where
C := C [ C is clopen and y C
Why does this suce? Well, then we get C O having empty intersection, so by com-
pactness we get C
1
, . . . , C
n
C such that C
1
C
n
(O) = . Then C
1
C
n
is
the desired clopen set C
y
. So we show that

C = y
Suppose, to get a contradiction, that

C has 2 or more elements in it. Then, it is


disconnected and we may introduce open sets A

, B

such that (

C) A

, (

C) B

is a
nontrivial partition of

C. As

C is closed, and A

and B

are closed, we may introduce


closed sets A and B such that A B = and A B =

C. As X is compact hausdor, it
is normal by Proposition 5 and so we get open O
A
and O
B
such that A O
A
, B O
B
and
O
A
O
B
= . Without loss of generality, assume that y A. Let F = (O
A
O
B
), a closed
set.
Since F (

C) = , C is consistent, C is closed under nite intersections, and Y is


compact, there is some C C such that C F = . Of course, O
A
O
B
= . Thus,
(C O
A
) (F O
B
) =
I.e., C O
A
O
A
. Since C is closed, this shows that C O
A
= C O
A
. I.e., C O
A
is
closed. It, being the intersection of two open sets, is also open. So C O
A
is a clopen set
containing y, yet it is disjoint from nonempty B

C, a contradiction.
Path-connected A path in a space X is a continuous function from the unit interval
I = [0, 1] into X. X is said to be path-connected if any two points can be joined by a path.
I.e., for any x
1
, x
2
X there is a path f : [0, 1] X such that f(0) = x
1
and f(1) = x
2
.
Let x X. X is path-connected i for every y X there is a path joining x and
y, because we can put two paths together piecewise. From this observation we get that
path-connectedness implies connectedness. As, I is connected, and continuous functions
preserve connectedness, we get that every point y X is in the component of x, and so X
is connected.
Its not in general true that every connected space is path-connected.
7 Stone Duality
There is a close connection between the category of boolean algebras with boolean homo-
morphism (BOOL) and the category of stone spaces with continuous maps (STON). A
stone space is a topological space which is compact, hausdor, and has a base of clopen sets.
We shall dene contravariant functors S: BOOL STON and B: STON BOOL
such that S and B are essentially inverses of each other. I.e., well show that SB is naturally
isomorphic to I
STON
and BS is naturally isomorphic to I
BOOL
. In particular, we get that
14
for any boolean algebra B, S(B) is a stone space, every stone space is homeomorphic to a
topology that arises in this way.
First we dene S: BOOL STON. Given a boolean algebra B, we let the underlying
set of S(B) consist of the ultralters on B, and then we declare the basic open sets to be
U
b
:= U [ b U (b B)
It is straightforward to check that this yields a stone space, with compactness following from
the ultralter lemma.
Next, given a boolean homomorphism h: B
1
B
2
, we shall dene a continuous map from
S(B
2
) to S(B
1
) (recall that our functors are to be contravariant). S(h)(U) := h
1
(U). This
is a well-dened map, because, as can be checked, h
1
(U) is an ultralter of B
1
. Further,
S(h) is continuous. It suces to show that [S(h)]
1
(U
b
) is open for each b B
1
. But it can
be checked that [S(h)]
1
(U
b
) = U
h(b)
Now we may dene the functor B: STON BOOL. Given a stone space X, we may
form the boolean algebra B(X) consisting of all clopen subsets of X. Being a subalgebra of
a powerset algebra, B(X) is indeed a boolean algebra.
Next, given f : X
1
X
2
continuous, we dene a boolean homomorphism B(f): B(X
2
)
B(X
1
) as follows. B(f)(C) := f
1
(C). Again, it is easy to check that this yields a well-dened
boolean homomorphism which sends a clopen of X
2
to a clopen of X
1
.
Now we verify that S and B give a duality of categories. We show rst that BS is naturally
isomorphic to I
BOOL
. For each boolean algebra B, we dene an isomorphism
B
: B BS(B)
such that for any boolean homomorphism h: B
1
B
2
, we have
B
2
h = BS(h)
B
1
. I.e., not
only is B isomorphic to BS(B), but morphisms into and out of B correspond appropriately
to morphisms into and out of BS(B).
The map
B
that works here is dened by:
B
(b) := U
b
. It is straightforward to check
that this is well-dened, is a bijective homomorphism, and obeys the naturality diagram.
Now, to show that SB is naturally isomorphic to I
STON
, we dene for every stone space
X an isomorphism
X
: X SB(X) such that for any continuous map f : X Y we have

Y
f = SB(f)
X
. The map that works is dened by
X
(x) := C
x
where C
x
consists of
all the clopen sets containing x.
8 Completeness and the Baire Category Theorem
A sequence in a metric space is said to be cauchy if the distances between terms become
small. Formally, x
n
is cauchy means that for every > 0 there is some N N such that for
all n, m N, d(x
n
, x
m
) < .
Recall that x
n
converges to x i for every r > 0, x
n
is eventually in B(x, r). It follows
that every convergent sequence is cauchy, since if x
n
, x
m
B(x, r), then d(x
n
, x
m
) < 2r.
However, the converse is not true. For example, consider a sequence in Q which converges
to

2. This is cauchy, but not convergent (in Q). Or, similarly,

n
i=1
p
n
converges to
1
1p
in
the p-adic metric, but if we remove the limit, say just consider the integers, then it remains
cauchy but is no longer convergent.
15
A metric space is called complete if every cauchy sequence converges. For example, Q is
not complete, but R is.
Completeness is a property of metrics, not topologies. There are topologically equivalent
metrics such that one is complete and the other not. For example, for x, y R set
d

(x, y) =

x
1 +[x[

y
1 +[y[

One can check that the map x


x
1+|x|
is a homeomorphism of R and (1, 1) (with the sub-
space topology). d

is dened on R so that this map is an isometry of (R, d

) and ((1, 1), d)


where d denotes the usual metric. It follows that (R, d

) is a well-dened metric space


homeomorphic to (R, d). Further, (1, 2, 3, . . .) is cauchy yet not convergent in (R, d

).
Now we turn our attention to the Baire category theorem. A subset Y of a topological
space X is said to be dense if

Y = X. Because the closure of Y may be described as the
set of all elements of X whose open neighborhoods always intersect Y , Y is dense i every
nonempty open set U of X intersects Y .
Theorem 8 (Baire). Let X be a complete metric space, and let O
n
(n ) be a countable
collection of dense open sets. Then

n
O
n
is dense.
Proof. Let U be a nonempty open set. We show that U and

n
O
n
contain a common
element. Since O
0
is dense, we may introduce an element x
0
U O
0
. Since U O
0
is
open, we may introduce a positive real number r such that B(x
0
, r) U O
0
. Then let
r
0
< r. Set B
0
:= B(x
0
, r
0
) and note B
0
U O
0
. Next, we note that B
0
is a nonempty
open set, and as O
1
is dense, we may introduce x
1
B
0
O
1
. Since B
0
O
1
is open, we
may introduce an r

> 0 such that B(x


1
, r

) B
0
O
1
. Choose r
1
so that r
1
< r

,
r
0
2
. Then
set B
1
:= B(x
1
, r
1
). Since r
1
< r

, we have B
1
B
0
O
1
. Similarly, we may dene (x
2
, r
2
)
such that r
2
<
r
1
2
and B
2
B
1
O
2
. By induction we obtain, for each n , elements x
n
and positive real numbers r
n
such that
1. r
n
converges to 0
2. B
n
:= B(x
n
, r
n
) is a decreasing sequence of sets
3. B
n
O
n
for each n
4. B
0
U
We claim that x
n
is a cauchy sequence. Let > 0 be given. Introduce N such that
2r
N
< . Then n, m N implies that x
n
, x
m
are both in B
N
. Thus, the distance between x
n
and x
m
is less than 2r
N
, the diameter of the ball B
N
. Since X is complete, we may introduce
a limit x of this sequence. We claim that x U and x

n
O
n
. Since x
n
is a sequence in
B
0
and B
0
U, it follows that x U. (For metric spaces we have that y

Y i there is a
sequence in Y converging to y.) Similarly, x
m
(m n) is a sequence in B
n
for each n ,
and since B
n
O
n
, we see that x O
n
for each n .
16
An application of this theorem is to show that there are many continuous, nowhere
dierentiable functions.
Note that the conclusion of the theorem only involves topological properties, though one
of the hypotheses is that X is complete, and this is a metric notion. If X is any topological
space with some associated metric that is complete, then the countable union of open dense
sets is dense.
Alternative Terminology We may also express this theorem in dierent terminology. A
subset Y of a space X is called nowhere-dense if the interior of the closure of Y is empty.
I.e., int(

Y ) = . This is equivalent to saying that the closure of Y contains some nonempty


open set. If we use Y is dense in Z to mean Z

Y , then Y is nowhere-dense is equivalent
to saying that Y is not dense in O for any nonempty open set O. A nowhere-dense set should
intuitively be thought of as small. For example, a nite subset of R is nowhere-dense.
Since the closure of the closure of Y is simply the closure of Y , it follows that Y is
nowhere-dense i

Y is nowhere-dense. If Y is closed, then Y is nowhere-dense i int Y =
i Y = X, i.e the complement of Y is dense. Thus, in general, Y is nowhere-dense i

Y is
nowhere-dense i the complement of

Y is dense i the exterior of Y is dense i the interior
of Y is dense.
Subsets of nowhere-dense sets are nowhere-dense, and nite unions of nowhere-dense sets
are nowhere-dense. I.e., the nowhere-dense sets form an ideal. But its not in general true
that countable unions of nowhere-dense sets are nowhere-dense. For example, the rationals
are a countable union of singletons of R, but the rationals are not nowhere-dense. In fact,
the rationals are dense and so their closure contains every open set of R.
A set is said to be meager if it is the countable union of nowhere-dense sets. We may
also think of meager sets as small, though of course they sometimes get rather large in the
sense that they may be dense. Another phrase used to express that Y is meager is Y is of
rst category. If Y is not meager, then Y is said to be of second category.
A set is said to be comeager if its complement is meager. In view of our previous
characterization of nowhere-dense sets as sets whose complements have dense interiors, we
note that Y is comeager i Y is the countable intersection of sets whose interior is dense.
A topological space is said to be a Baire space, if the conclusion of the theorem above
holds, i.e., the intersection of countably many dense open sets is dense. Using the terminology
just given, we may express this condition in dierent ways.
Proposition 9. The following conditions on a topological space X are equivalent:
1. X is a Baire space, i.e. the intersection of countably many dense open sets is dense.
2. Every meager set has empty interior.
3. Every comeager set is dense.
17
Proof. Suppose every countable intersection of dense open sets is dense. Let Y be a meager
set. Introduce nowhere-dense sets A
i
such that Y =

i
A
i
. Of course
Y
_
i
A
i
So it suces to show that

_
i
A
i
=

i
A
i
is dense. But each A
i
is a closed nowhere-dense set, and so its complement A
i
is an open
dense set.
Now suppose that every meager set has empty interior. We show that the countable
intersection of dense open sets is dense. Let

i
O
i
be a countable intersection of dense
open sets. It suces to show that

i
O
i
=
_
i
O
i
has empty interior. But each O
i
is a closed set with empty interior, and hence nowhere-
dense. Thus, the displayed set is meager and by assumption has empty interior.
Finally we note the latter two conditions are equivalent. The comeager sets are the
complements of the meager sets, and a set is dense i its complement has empty interior.
We may rephrase the Baire category theorem to assert that any completely metrizable
space is a Baire space (i.e. a space where any of the three equivalent conditions above hold).
For example, any second countable compact hausdor space is metrizable by Urysohns
metrization theorem, and any such metric must be complete, because the space is compact
(the Heine-Borel theorem puts compact equivalent to complete plus totally bounded). So
any second countable compact hausdor space is a Baire space. In particular, a topological
space whose basic sets are the sentences of a countable rst order language is a Baire space.
The Baire cateogry theorem is analogous to the omitting types theorem, because both assert
the existence of a simultaneous solution to countably many problems.
9 Kuratowski-Ulam Theorem
We now turn to the Kuratowski-Ulam theorem, which is analogous to Fubinis theorem of
measure theory. Let X and Y be spaces and let E be a subset of X Y . For x X, we
may form the x-section of E, denoted E
x
, dened by
E
x
:= y Y [ (x, y) E
I.e., if we imagine X as a horizontal axis and Y as a vertical axis, then E
x
is the vertical
slice of E at x. E and its sections of course bear certain relationships. For example, if
18
E =

iI
E
i
, then E
x
=

iI
(E
i
)
x
. Also, if E is open the so too is E
x
. To see this, let
E =

iI
(U
i
V
i
) where, for each i, U
i
is an open set of X and V
i
is an open set of Y . Then
E
x
=
_
xU
i
V
i
which is open, being the union of open sets.
For the theorem, we also need the notion of a set with the Baire property. A subset Y of
a space X has the Baire property means that there is some open set O and meager set M
such that Y = OM, where as usual denotes the symmetric dierence. If we think of
meager sets as small, then this denition says that Y is almost open. However, almost
open is equivalent to almost closed. In fact, the sets with the Baire property form a
-algebra on X in the sense that this collection is closed under complements and under
countable union (and countable intersection). Its the smallest -algebra that contains all
the open and meager sets.
Theorem 10 (Kuratowski-Ulam). Let X and Y be spaces with Y second countable. Then
the following things are true:
1. If E is meager, then x X [ E
x
is meager is comeager.
2. If E has the Baire property, and x X [ E
x
is meager is comeager, then E is
meager.
The second part is a converse of the rst part, but we need the additional assumption
that E has the Baire property. The rst part may be interpreted as saying that if E is
small, then most of its sections are small too. Similarly, the second part can be interpreted
as saying that if most of its sections are small, then E is small too. If we assume that X is
second countable, then of course the theorem also holds for sections E
y
.
Proof. We consider the rst part rst. We note it suces to show that if E is nowhere-dense,
then x X [ E
x
is nowhere-dense is comeager. To see that this suces, let E be meager
and so write E as the countable union of nowhere-dense sets E =

i
E
i
. Since comeager
sets are closed under countable intersection, we have that

i
x X [ (E
i
)
x
is nowhere-dense
is comeager. Since comeager sets are closed under superset, it suces to show that

i
x X [ (E
i
)
x
is nowhere-dense x X [ E
x
is meager
Well, suppose that x is such that for each i , (E
i
)
x
is nowhere-dense. As we noted
previously, E
x
=

i
(E
i
)
x
, and so E
x
is meager.
19
So let E be nowhere-dense, and we show that x X [ E
x
is nowhere-dense is comeager.
Let V
n
(n ) be a countable base for Y , except dont include . Of course, G :=

E, the
complement of the closure of E, is dense and open. For each n , we dene
G
n
:= p
X
(G (X V
n
))
I.e.,
G
n
= x X [ y V
n
[(x, y) G]
Each G
n
is an open set of X, because the projection map p
X
is open, and G (X V
n
) is
a nite intersection of open sets. In fact, each G
n
is also dense. To see this, let U be any
nonempty open set of X and we show G
n
U ,= . Since U V
n
is a nonempty open set of
X Y , and G is dense, we may introduce (x, y) G (U V
n
). Then x G
n
U.
Since each G
n
has dense interior (the interior being the same as G
n
), we see that

n
G
n
is comeager. So it suces to show that

n
G
n
x X [ E
x
is nowhere-dense
So let x G
n
for each n . It follows that the section G
x
intersects V
n
for every n .
So G
x
is dense (in Y ). G
x
is also open, being the section of an open set G. Thus G
x
is a
nowhere-dense set. Also, since (x, y) E implies (x, y)

E, and as G =

E, this in turn
implies (x, y) , G, we see that E
x
G
x
, and so it is nowhere-dense too.
Now we turn to the second part. Here we assume that E has the Baire property, and that
x X [ E
x
is meager is comeager. Assume, to get a contradiction, that E is not meager.
Then we may write E = OM where M is meager and O is open, and further O is not
meager, lest E be meager too. O is of course a union of basic open sets U V . Lemma 11
shows that not all of these basic open set are meager. Thus, we may introduce open sets
U X and V Y such that U V O and U V is not meager. From Lemma 12 we get
that both U and V are not meager.
The contradiction we want to get is that x X [ E
x
is not meager is not meager. By
the rst part of our theorem, we get that x X [ M
x
is not meager is meager. Since a non-
meager set minus a meager set is still non-meager, we get that U x X [ M
x
is meager
is non-meager. Since a superset of a non-meager is still non-meager, it suces to show that
U x X [ M
x
is meager x X [ E
x
is not meager
To see this, let x U such that M
x
is meager. Then V M
x
is nonmeager since its a
nonmeager set minus a meager set. So it suces to show that E
x
V M
x
. Well, let
y V M
x
. Then (x, y) U V O, and (x, y) , M. Thus, (x, y) OM = E. I.e.,
y E
x
.
Lemma 11 (Banach Category Theorem). If O
i
(i I) is a collection (not necessarily
countable) of meager open sets of a space X, then

iI
O
i
is also meager.
20
Proof. We may introduce a maximal family of disjoint open sets F = U

such that each


U

is contained in some O
i
. Such a maximal family exists because the union of chains of
such families is itself such a family.
Write G for

iI
O
i
. We claim that

G (

F) is a closed, nowhere-dense set. It is


clearly closed, being the intersection of closed sets. To show its nowhere-dense, we show
that it doesnt contain any nonempty open set. Suppose, to get a contradiction, that U

G(

F) with U nonempty open. Then, as U G ,= we may introduce an O


i
such that
U O
i
is nonempty. Then U O
i
is a nonempty open set which is a subset of O
i
. Since
U O
i
(

F), we see that U O


i
is disjoint with every element of F. Thus, F U O
i

violates the maximality of F.


Of course
G (

G (
_
F))
_
F
So to show that G is meager, it suces to show that the set on the right side is meager. Since

G(

F) is nowhere-dense, it in turn suces to show that

F is meager, i.e. expressible


as the countable union of nowhere-dense sets.
Each U

in F is a subset of some meager O


i
, and hence meager. So, for each , introduce
nowhere-dense sets A
,n
(n ) such that
U

=
_
n
A
,n
Let B
n
:=

A
,n
for each n . Since
_
F =
_

_
n
A
,n
=
_
n
B
n
it suces to show that each B
n
is nowhere-dense.
B
n
is nowhere-dense i for every nonempty open set U there is some nonempty open set
V such that V U B
n
. So let U be a nonempty open set. If U B
n
, then we may let
V = U. So assume that U B
n
,= 0. Then we may introduce an such that U A
,n
,= .
In particular, U U

is a nonempty open set. Since A


,n
is nowhere-dense, we may introduce
a nonempty open set V such that V (U U

) A
,n
. Then of course V U. Also,
V B
n
because if

,= , then A

,n
U

= (F consists of disjoint sets).


Lemma 12. Let U be a subset of X and V be a subset of Y . If either of U or V is meager,
then U V is meager too.
Proof. First we observe that if G is an open dense subset of X, then G Y is an open
dense subset of X Y . Of course G Y is open. To see that its dense we may show
that the intersection of G Y with any nonempty basic open set O
1
O
2
is nonempty.
Well, let (x, y) O
1
O
2
. Then, as G is dense, we get an element x
1
O
1
G. So
(x
1
, y) (O
1
O
2
) (GY ).
Now, we use this fact to show that if A is nowhere-dense in X, then A B is nowhere-
dense in X Y for any subset B of Y . If A is nowhere-dense, then int(A) is a dense open
21
set of X. By our previous observation, [int(A)] Y is an open dense subset of X Y . So,
to show that int((A B)) is dense, it suces to show that
[int(A)] Y int((A B))
Since [int(A)] Y is open, it suces to show that this set is contained in (AB). So let
(x, y) [int(A)] Y . In particular, x A, so (x, y) , A B, as desired.
Finally, we show that if U is meager then U V is meager (if V is meager the argument
is the same). We write U =

i
A
i
where the A
i
are nowhere-dense. Then by our previous
observation, we get that each A
i
V is nowhere-dense in X Y . Thus,
U V = (
_
i
A
i
) V =
_
i
(A
i
V )
is meager.
22

You might also like