Applications of Baire Category Theorem

S. Kumaresan
School of Math. and Stat.
University of Hyderabad
Hyderabad 500046
kumaresa@gmail.com

1 Baire Category Theorem

Definition 1. A subset A ⊂ X of a topological space is said to be nowhere dense in X,

if given any nonempty open set U , we can find a nonempty open subset V ⊂ U such that
A ∩ V = ∅.
This definition is equivalent to the standard one found in all text-books: A is nowhere
dense in X iff the interior of the closure of A in X is empty.
I prefer the first one, as it gives us a better geometric intuition as it uses only the primitive
concepts of topology.

Example 2. Let V be any proper vector subspace of Rn . Then V is nowhere dense in Rn .

This is a typical example of a nowhere dense set.
More generally, let X be a normed linear space. Let V be any proper vector subspace of
X. Then V is nowhere dense in X.

We shall give the formulation of Baire category theorem in a form which will be more
useful than the one which uses the notion of category.

Theorem 3 (Baire Category Theorem). Let (X, d) be a complete metric space.

(1) Let Un be open dense subsets of X, for n ∈ N. Then ∩n Un is dense in X.
(2) Let Fn be nonempty closed subsets of X such that X = ∪n Fn . Then at least one of Fn ’s
has nonempty interior. In other words, a complete metric space cannot be a countable union
of nowhere dense closed subsets.

Proof. We first observe that both the statements are equivalent. For, G is open and dense iff
its complement F := X \ G is closed and nowhere dense. Hence any one of them follows from
the other by taking complements. So, we confine ourselves to proving the first.
Let U := ∩n Un . We have to prove that U is dense in X. Let x ∈ X and r > 0 be
given. We need to show that B(x, r) ∩ U 6= ∅. Since U1 is dense and B(x, r) is open there
exists x1 ∈ B(x, r) ∩ U1 . Since B(x, r) ∩ U1 is open, there exists r1 such that 0 < r1 < 1/2
and B[x1 , r1 ] ⊂ B(x, r) ∩ U1 . We repeat this argument for the open set B(x1 , r1 ) and the

1
dense set U2 to get x2 ∈ B(x1 , r1 ) ∩ U2 . Again, we can find r2 such that 0 < r2 < 2−2 and
B[x2 , r2 ] ⊂ B(x1 , r1 ) ∩ U2 . Proceeding this way, we get for each n ∈ N, xn and rn with the
properties
B[xn , rn ] ⊂ B(xn−1 , rn−1 ) ∩ Un and 0 < rn < 2−n .
Clearly, the sequence (xn ) is Cauchy: if m ≤ n,
n
X
d(xm , xn ) ≤ d(xn , xn−1 ) + · · · + d(xm+1 , xm ) ≤ 2−k .
k=m

2−k is convergent, it follows that (xn ) is Cauchy.

P
Since k

Since X is complete, there exists x0 ∈ X such that xn → x0 . Since x0 is the limit of the
sequence (xn )n≥k in the closed set B[xk , rk ], we deduce that x0 ∈ B[xk , rk ] ⊂ B(xk−1 , rk−1 ) ∩
Uk for all k. In particular, x0 ∈ B(x, r) ∩ Uk .

Remark 4. The importance of our formulation is this. The first statement tells us of a
typical way in which Baire category can be used. Imagine that we are on the look-out for
an element x ∈ X with some specific properties. Further assume that the sets of elements
which have properties “arbitrarily close” to the one desired are dense open sets in X. Then
the result says that there exists at least one element with the desired property. Thus the first
formulation is useful when we are interested in the existence problems. This vague way of
remembering is well-illustrated in some of the applications below. See especially the existence
of everywhere continuous nowhere differentiable function.
The second formulation says that X cannot be a countable union of “hollow” sets. A
typical application: Rn cannot be the union of a countable collection of lower dimensional
subspaces. Another instance: a complete normed linear space cannot be countable dimen-
sional. See Prop. 8 below.

2 Uniform Boundedness Principle

Theorem 5. Let X be a complete metric space and F a family of continuous real valued
functions on X. Assume that for all x ∈ X, there exists Cx > 0 such that |f (x)| ≤ Cx for all
f ∈ F. Then there exists a non-empty open set U ⊆ X and a constant C such that |f (x)| ≤ C
for all f ∈ F and x ∈ U .

Proof. For each f ∈ F and n ∈ N, let En,f = {x ∈ X | |f (x)| ≤ n}. En,f is closed since f is
continuous. Let En = ∩f ∈F En,f . Then En is closed and ∪En = X, since if x ∈ X, x ∈ En if
n > Cx . By Baire category, not all En ’s are nowhere dense. Hence En = En must contain a
non-empty open set U for some n. Take C = n.

Theorem 6. . Let X be a Banach space and C a closed, convex, symmetric subset of X such
that ∪n nC = X. Then C is a neighborhood of 0.

Proof. All nC are closed since multiplication by a non-zero scalar is a homeomorphism. Hence
there exists N > 0 such that N C has a non-empty interior, by Baire Category Theorem
and hence C has a non-empty interior (multiply by N1 !). Hence there exists x ∈ C and a

2
r > 0 such that x + r(B(0, 1)) ⊆ C. Since C is symmetric, −x ∈ C, so that by convexity,
1 1 1 r
2 r(B(0, 1)) = 2 (−x) + 2 (x + B(0, 1)) ⊆ C. That is, C contains the neighborhood B(0, 2 ) of
0.

Remark 7. This underlies the proof of the open mapping theorem.

Proposition 8. Let X be an infinite dimensional complete normed linear space. Then X
cannot be countable dimensional.

Proof. Let, if possible, {en : n ∈ N} be a (Hamel/algebraic) basis of X. This means that any
vector x ∈ X is a finite linear combination of en ’s. If we let Fn stand for the vector subspace
spanned by {ek |1 ≤ k ≤ n}, then Fn is finite dimensional. It is well-known that all norms on
a finite dimensional vector space are equivalent so that any finite dimensional normed linear
space is necessarily complete. In view of this we conclude that each Fn is closed in X. By
Eg. 2, Fn is closed for each n. By hypothesis, the complete metric space X is the union of
the countable family {Fn } of nowhere dense closed sets, a contradiction.

Theorem 9. If fn ∈ C[a, b] are such that fn (t) → f (t) for each t ∈ [a, b], then the set of
points of continuity of f is dense in [a, b].

We first prove a lemma.

Lemma 10. To each closed subinterval I of [a, b] and every ε > 0, there exists a closed
subinterval J ⊆ I such that |f (t1 ) − f (t2 )| ≤ ε for all t1 , t2 ∈ J.

Proof. Let Fn := {t ∈ I | |fn (t) − fm (t)| ≤ 3ε for all m ≥ n}. Obviously, every Fn is closed
and I = ∪Fn . By Baire category theorem, at least one of them, say Fk contains a closed
(why?) interval I ′ . For all t ∈ I ′ and m ≥ k we have
ε
|fk (t) − fm (t)| ≤
3
and hence also |fk (t) − f (t)| ≤ 3ε . Since fk is uniformly continuous on I ′ , there exists a closed
subinterval J of I ′ so that
ε
|fk (s) − fk (t)| ≤ for all s, t ∈ J.
3
For such points s, t ∈ J, we have

|f (s) − f (t)| ≤ |f (s) − fk (t)| + |fk (s) − fk (t)| + |fk (t) − f (t)|
ε
≤ 3 =ε
3

Proof. From the lemma, it follows that for every closed subinterval I of [a, b], we can construct
a sequence of closed intervals In := [an , bn ] such that In ⊆ I, an < an+1 ≤ bn+1 < bn ,
bn − an < n1 and |f (s) − f (t)| ≤ n1 for s, t ∈ In . By Cantor’s intersection (“nested interval”)
theorem, there exists a common point x ∈ ∩In . At this point f is obviously continuous.
Thus, if t is an arbitrary point of [a, b] and ε > 0, then [t − ε, t + ε] contains a point x at
which f is continuous. Hence the set of points of continuity of f is dense in [a, b].

3
Ex. 11. An amusing exercise: Let (xn ) be any sequence of real numbers. Show that the set
{x ∈ R : x 6= xn , n ∈ N} is dense in R. Hence conclude that R is uncountable.
Show that Q cannot be written as the intersection of a countable family of open subsets
of R.

The next result is a beautiful application of Baire’s theorem which uses both the versions!
To put it in perspective, recall that the pointwise limit of a sequence of continuous functions
need not be continuous while the uniform limit is. However, the poinwise limit cannot be too
wild.

Theorem 12. Let X be a complete metric space. Let fn : X → R be a sequence of continuous

functions. Assume that there exists a function f : X → R such that fn (x) converges to f (x)
for each x ∈ X. Then there exists a dense subset D of X such that each point of D is a point
of continuity of f .

Proof. Fix ε > 0. Define, for each k ∈ N,

Ek (ε) := {x ∈ X : |fn (x) − fm (x)| ≤ 1/k, for all m, n ≥ n}.

Then we claim that Ek (ε) is closed for each k.

Reason: Fix m, n ≥ k. Then the set Ekm,n (ε) := {x ∈ X : |fn (x) − fm (x)| ≤ 1/k} is a
closed subset of X, since |fn − fm | is continuous. Now, since Ek (ε) = ∩m,n≥k Ekm,n (ε),
the claim follows.

It is easy to show that X = ∪k Ek (ε).

Reason: Let x0 ∈ X. Since fn (x0 ) → f (x0 ), the sequence (fn (x0 )) is Cauchy. Hence for
the given ε > 0, there exists k0 such that for m, n ≥ k0 , we have |fm (x0 ) − fn (x0 )| ≤ 1/k0 .
Hence we conclude that x0 ∈ Ek0 (ε).

Since Xis a complete metric space, at least one of Ek (ε) should have nonempty interior.
Let Uε := ∪k Int (Ek (ε)). Then Uε is a nonempty open subset of X.
Let Un := U1/n . We claim that each Un is dense in X.

Reason: It is enough if we show that every closed ball B := B[x, r] meets Un nontrivially.
(Why?)

Reason: To show a set A is dense in a metric space, it suffices to show that

A ∩ B(x, r) 6= ∅ for any x ∈ Xand r > 0. Assume that A ∩ B[z, ρ] 6= ∅ for any
z ∈ X and ρ > 0. Then given any B(x, r), we may take z = x and ρ = r/2.
Then ∅ 6= A ∩ B[x, ρ] ⊂ A ∩ B(x, r).

Observe that the closed set (and hence a complete metric space) B is the union of a
countable family of closed sets: B = ∪n (B ∩ Ek (1/n)). By Baire, at least one of them has
nonempty interior, say, Int (B ∩ Ek (1/n)) 6= ∅. Since Int (B ∩ Ek (1/n)) ⊂ B ∩ Int Ek (1/n),
it follows that B[x, r] ∩ Un 6= ∅ and hence the claim is proved.

4
 Let D := ∩n Un . By Baire, D is dense in X. We claim that every x ∈ D is a point of
continuity of f .

Reason: Fix p ∈ D. Let ε > 0 be given. Choose N ≫ 0 such that 1/N < ε. Since p ∈ D,
p ∈ UN and hence there exists k ∈ N such that p ∈ Int (Ek (1/N ). By continuity of fk at
p, there exists an open neighbourhood V of p contained in Int Ek (1/N ) such that

|fk (x) − fk (p)| < ε, for all x ∈ V. (1)

For x ∈ V , since V ⊂ Ek (1/N ), by the definition of Ek (ε)’s, we have

|fm (x) − fk (x)| ≤ 1/N, for all m ≥ k. (2)

Letting m → ∞ in the above equation, we obtain

|f (x) − fk (x)| ≤ 1/N, for all x ∈ V. (3)

We are now ready for the kill. We claim that |f (x) − f (p)| < 3ε for x ∈ V .

|f (x) − f (p)| ≤ |f (x) − fk (x)| + |fk (x) − fk (p)| + |fk (p) − f (p)|
≤ 1/N + ε + 1/N
< 3ε.

This shows that f is continuous at every point of D.

Theorem 13. There exists a continuous function on [0, 1] which is not differentiable at any
point.

Proof. For each positive integer n, we let

  
  f (t+h)−f (t) 
Cn := f ∈ C[0, 1]]  

h  ≤ n for some t and all h with t + h ∈ [0, 1] .

We shall show that Cn is nowhere dense for each n. Since C[0, 1] is a complete metric space,
it will then follow that ∪∞ i=1 Cn 6= C[0, 1]. Hence there exists f ∈ C[0, 1] such that f ∈ / Cn for
any n.
 
This f is the required one. For, f ∈ / Cn means that  f (t+h)−f (t) 
 > n for all t ∈ [0, 1] and

h
some h (depending on t and n). Note that for each fixed t, h → 0 as n → ∞: Because, given
ε > 0, the difference quotient  f (t+h)−f (t) 
 is bounded as a function of h for |h| ≥ ε. Thus

h
 
 f (t+h)−f (t) 
lim suph→0  h  = ∞, and the derivative of f at t fails to exist for each t ∈ [0, 1].

To prove that Cn is nowhere dense, we must show that Cn contains no non-empty open
set. First we show that Cn is closed, i.e., Cn = Cn .
Since C[0, 1] is a metric space, it is first countable and hence it suffices to show that if
{fk } ⊆ Cn , with lim fk = f ∈ C[0, 1], then f ∈ Cn . Now, fk ∈ Cn implies that there exists
tk ∈ [0, 1] such that  
 fk (tk + h) − fk (tk ) 
  ≤ n, for all h.
 h 

5
Since [0, 1] is compact, the sequence tk has a convergent subsequence. We call this subsequence
again by tk and let t0 = lim tk . Then
     
 f (t0 + h) − f (t0 ) 
 ≤  f (t0 + h) − f (tk + h)  +  f (tk + h) − fk (tk + h) 
   

 h   h   h 
   
 fk (tk + h) − fk (tk )   fk (tk ) − f (tk ) 
+  + 
h   h 
 
 f (tk + h) − f (t0 ) 
+  
h 
= (1) + (2) + (3) + (4) + (5)

Now, fix h. For any ε > 0, if k is large enough, (1) and (5) are smaller than ε, since f is
continuous, and tk → t0 . (2) and (4) are smaller than ε, since fk converges uniformly to f .
(3) ≤ n. Hence, we get
 
 f (t0 + h) − f (t0 ) 
  ≤ n + 4ε for any ε > 0
 h 

so that  
 f (t0 + h) − f (t0 ) 
 ≤n
 h 
and hence f ∈ Cn . Thus each Cn is closed.
We now show that Cn is nowhere dense; that is, given any g ∈ Cn = Cn and any ε > 0,
there exists f ∈ C[0, 1] such that kf − g k < ε and f ∈
/ Cn .
Now a typical example of a function in C[0, 1] which is not in Cn is the “sawtooth” function.
For any n, we can find such a function, whose norm is less than or equal to any prescribed
ε > 0, and where the slope of each line segment is greater than n in absolute value. To
find such a function f ∈ / Cn , with an ε-distance from g, we need only construct a sawtooth
function close to g, as in Figure 1.
g

−ε

Figure 1: “Saw-tooth” Function

To construct such a function f , we use the uniform continuity of g to get a piecewise

linear function g1 with kg − g1 k < 2ε . For each linear piece of g1 we construct a saw-tooth
function.

6
3 A Property of Repeated Integrals

Theorem 14. Let f ∈ C[0, 1]. Let f1 be any integral of f and let fn+1 be any integral of fk ,
k ≥ 1. If some fk vanishes identically, so does f .

We need only differentiate fk+1 repeatedly. We now wish to prove a generalization.

Theorem 15. If, for each x, there is an integer k = k(x), such that fk (x) = 0, then f
vanishes identically.

Proof. Let Ek := {x ∈ [0, 1] | fk (x) = 0}. By hypothesis, any x ∈ [0, 1] lies in some Ek . Ek is
obviously closed. Hence by Baire category theorem, there exists some k for which Ek contains
an interval Ik . For this k, since fk is continuous and vanishes on Ek , we have fk (x) = 0 for
all x ∈ Ik .
If Ik 6= [0, 1], we repeat this argument with any complementary subinterval and so on. In
this way, we have f (x) = 0 for all points x of an everywhere dense set. Since f is continuous,
f = 0 on [0, 1].

An Alternate Proof. Let I = [0, 1]. Let J0 be a non-empty, closed subinterval such that
fk (x) = 0 for all x ∈ J0 . Then we have f ≡ 0 on J0 .
Let E := {x ∈ I | f (x) = 0} ⊇ I0 . E is closed. If E 6= [0, 1], then there exists an open
interval in J \ E and hence a non-empty closed subinterval J ⊆ I \ E. Note that f (x) 6= 0
for all x ∈ J.
Using the argument by replacing the pair (f, I) with (f, J), we conclude that there exists
a non-empty closed subinterval J0 ⊆ J such that fl (x) = 0 for all x ∈ J0 . Hence f ≡ 0 on J0 .
This contradicts the fact that J0 ⊆ J ⊆ I \ E and hence E = I.

4 A Characterization of Polynomials.

Let f on [0, 1] be such that its n-th derivative is 0. By repeated application of mean value
theorem, we see that f coincides on [0, 1] with a polynomial (of degree atmost n − 1). We
generalize this as follows:
Theorem 16. Let f be C ∞ . Let, for each x ∈ [0, 1], there be n(x) ∈ N such that f (n(x)) (x) =
0. Then f coincides with a polynomial.

Proof. Let En be the set of points x for which f (n) (x) = 0. Then [0, 1] = ∪En , by hypothesis.
As En is closed, by Baire’s theorem, there is a closed interval I in which some En is dense.
Since f (n) is continuous, f (n) ≡ 0 on I and f coincides with a polynomial on I. If I = [0, 1],
we repeat the same reasoning in the remaining part of [0, 1] and so on. In this way, we see
that there is a dense set of intervals in each of which f coincides with a polynomial. What
needs to be shown is that f coincides with the same polynomial in all the intervals.
We are going to apply Baire’s theorem again to the nowhere dense set H that is left when
we remove the interiors of the dense set of intervals from [0, 1].

7
 We show that H is perfect. H is closed, since it is obtained by removing a collection
of open intervals from a closed interval. If H is empty, then there was only one interval to
begin with and there is nothing to prove then. So we assume H is non-empty. If H were not
perfect, then H contains a point y that is not a limit point. y is then the common end point
of two intervals in each of which f coincides with some polynomial. Then if n exceeds the
degrees of these two polynomials, then f (n) (x) = 0 for x in both intervals, and also at the
common end point by continuity. Therefore, f coincides with a polynomial in the union of
the two intervals. But then y ∈
/ H, a contradiction.
It remains to be shown that H 6= ∅ leads to a contradiction.
If H 6= ∅, H is closed and hence a complete metric space. Now the J has intervals K
complementary to H, i.e., “discarded” in the first part of the proof. On such K, we have
f (m) (x) = 0, m = m(k), for all x ∈ K. If m ≤ k, f (n) (x) = 0, in K by differentiation. If
m > k, f (k) (x) = f (k+1) (x) = · · · = 0 at the end points of K, since these are points of H. If
we integrate f (m) repeatedly, we get f (k) (x) = 0 throughout K.
This reasoning is applicable to all intervals K which are complementary to H and lie in
J. Thus f (k) (x) = 0 for all points of J. But then J ∩ H = ∅. This contradiction shows that
H must be empty. This implies that there was only one interval which was “thrown out”.

Theorem 17. Let f : [0, ∞) → [0, ∞) be continuous. Assume that limn→∞ f (nx) = 0, for
any x ∈ [0, ∞). Then limx→∞ f (x) exists and is zero.

Proof. If the limit limn→∞ f (x) exists, one can easily show that it has to be 0. So, the main
thrust of the claim is that the limit exists. Let ǫ > 0 be given.
For each n, let En := {x ∈ [1, ∞) | |f (mx)| ≤ ǫ for all m ≥ n}. Then each En is closed
subset of [0, ∞). By hypothesis, [1, ∞) = ∪En . Since [1, ∞) is complete, Baire’s theorem
implies the existence of n for which En contains an open interval, say, (a, b), a < b. By our
definition of En , we have

|f (kx)| < ǫ for all x ∈ (a, b) and k ≥ n.

If k is sufficiently large, then (ka, kb) ∩ ((k + 1)a, (k + 1)b) 6= 0. That is, we choose k ≥ n
a
with (k + 1)a < kb or a < k(b − a) or k > max{n, b−a }. Then by hypothesis, on the union of
all such intervals, we have |f (x)| < ǫ. But then their union is of the form (ka, ∞). Then for
x ≥ ka, we have |f (x)| < ǫ. Hence limx→∞ f (x) exists and is 0.

Theorem 18. The set of continuous nowhere monotonic (≡ every where oscillating) functions
is dense in C[0, 1].

Proof. First notice that it is easy to give discontinuous nowhere monotonic functions.
Let {In } be the sequence of all closed intervals of [0, 1] with rational end points. Let Gn
denote the subset of C[0, 1] of functions which are not monotonic on In . (Read the description
of Gn carefully! We are not saying “nowhere monotonic on In ”.) The idea is to show that
Gn is both open and dense in C[0, 1] and apply Baire’s theorem.
Claim: Gn is open.

8
 a+b
0 a 2 b 1

Figure 2:

If f ∈ Gn , f is not monotonic on In . Hence there exists x < y < z with f (x) < f (y)
but f (z) < f (y) (or f (x) > f (y) and f (y) < f (z)). Now if we choose g with kf − g k <
1
2 min{f (y) − f (x), f (y) − f (z)}, we must have g must have similar properties. Thus all
elements g which are near to f lie in Gn .

Claim: Gn is dense, i.e., every neighborhood contains an f which is not monotonic on In .

It is enough to prove that Gn is dense in P L[0, 1], the space of piecewise linear functions,
since P L[0, 1] is dense in C[0, 1].
Let f ∈ P L[0, 1] \ Gn . Then f is piecewise linear and monotonic on In . We may assume
that f is non-decreasing on In .
Let [a, b] ⊆ In on which f is linear. If m is the slope of this line segment, the graph of f
on [a, b], we define 


 0 0≤x≤a
a < x ≤ a+b

2m(x − a)
h(x) = 2
a+b


 −2m(x − b) 2 < x ≤ b

0 b≤x<1

which is in P L[0, 1]. Let g = f + h. Then g( a+b

2 ) > g(a), g(b). But however, |g(x) − f (x)| =
|h(x)| < m(b − a). Hence if ǫ > 0 is given, the subinterval [a, b] may be chosen so that
m(b − a) < ǫ. Hence g ∈ B(f, ǫ). Thus Gn is dense in P L[0, 1] and hence in C[0, 1].
Since {Gn } are both open and dense in C[0, 1], G = ∩Gn is also dense in C[0, 1] and
therefore cannot be empty. However, if f ∈ G, then f ∈ Gn for all n and hence f is not
monotonic on any subinterval of [0, 1] with rational end points. The result follows.

Definition 19. Oscillation of a function:

Theorem 20. Let f : (a, b) → R be given. Then

(i) f is continuous at c ∈ (a, b) iff ωf (c) = 0.
(ii) For any ε > 0, the set {x ∈ (a, b) : ωf (x) < ε} is open.

9
Proof. Let f be continuous at c. Then given ε > 0, there exists δ > 0 such that for x ∈ J :=
(x − δ, x + δ), we have f (c) − ε < f (x) < f (c) + ε so that for x, y ∈ J, we have

f (y) − f (x) ≤ f (c) + ε − (f (c) − ε) = 2ε.

In particular, ωf (J) < 2ε. It follows that

0 ≤ ωf (c) ≤ inf{ωf (J) : where J corresponds to ε as above} = 0.

Conversely, if ωf (c) = 0, then given ε > 0, there exists an open interval J ∋ c such that
ωf (J) < ε. Choose δ > 0 such that (c − δ, c + δ) ⊂ J. In particular, for x ∈ (c − δ, c + δ), we
have |f (x) − f (c)| < ε. Thus, f is continuous at c. This proves (i).
To prove (ii), let ε > 0 be given. Let c be such that ωf (c) < ε. Thus, there exists an open
interval J ∋ c such that ωf (J) < ε. Thus for any x ∈ J, we have ωf (x) ≤ ωf (J) < ε. In
other words, c ∈ J ⊂ {y : ωf (y) < ε}. Thus c is an interior point of the set under question
and so the set is open.

Theorem 21. There exists no function f : (0, 1) → R which is continuous at rationals and
discontinuous at irrationals.

Proof. Assume such a function f exists. Consider the sets

Gn := {x ∈ (0, 1) : ωf (x) < 1/n}.

These sets are open in (0, 1) by the last lemma. Since (0, 1) is open in [0, 1], Gn is open in
[0, 1]. Since it contains Q ∩ (0, 1), it is dense in [0, 1]. Hence its complement Fn := [0, 1] \ Gn
in [0, 1] is nowhere dense and closed. Since ∩n Gn = (0, 1) ∩ Q, it follows that ∪n Fn is the set
of irrationals union {0, 1}. If we enumerate the set (0, 1) ∩ Q as {rn : n ∈ N}, it follows that

[0, 1] = ∪n Fn ∪ {r1 } ∪ · · · {rn } ∪ · · · ,

is a countable union of nowhere dense sets. This contradicts Baire Category theorem.

Theorem 22 (Baire category theorem for locally compact spaces). Let X be a locally compact
hausdorff space. Let (Un ) be a sequence of open dense sets in X. Then ∩n Un is dense in X.

Proof. Let G be a nonempty open set in X. We need to prove that there exists x ∈ G such
that x ∈ Un for all n. The strategy is to mimic the proof in the case of metric spaces replacing
open balls by the existence of open sets V such that V is compact and x ∈ V ⊂ V ⊂ U for any
given open set U and x ∈ U and then invoking Cantor intersection theorem for a decreasing
sequence of compact sets.
Since G is a nonempty open set and U1 is dense, there exists x1 ∈ G ∩ U1 . Since G ∩ U1 is
open, x ∈ G ∩ U1 and X is locally compact hausdorff space, there exists an open set V1 such
that x ∈ V1 , V 1 is compact and V 1 ⊂ G ∩ U1 . Assume, by way of induction, that we have
chosen xi , Vi ∋ xi , V i is compact and that xi ∈ Vi ⊂ V i ⊂ Vi−1 ∩ Ui , for 1 ≤ i ≤ n.
Now given a nonempty open set Vn , since Vn ∩ Un+1 is nonempty, there exists xn+1 ∈
Vn ∩ Un+1 . Since X is locally compact and hausdorff, there exists an open set Vn+1 ∋ xn+1

10
such that V n+1 is compact and xn+1 ∈ Vn+1 ⊂ V n+1 ⊂ Vn ∩ Un+1 . Let Kn := V n . Thus we
have a decreasing sequence (Kn ) of nonempty compact subsets. Hence by Cantor intersection
theorem, there exists x ∈ ∩n Kn . Since x ∈ Kn = V n ⊂ Un , it follows that x ∈ ∩Un . Also,
x ∈ K1 ⊂ U .

11