10 Compactness in function spaces: Ascoli-Arzela

theorem
Recall that if (X, d) is a compact metric space then the space C(X) is a vector
space consisting of all continuous function f : X R. The space C(X) is
equipped with the norm "f " := max{|f (x)| | x }. The norm induces the metric

(f, g) = "f g" = max{|f (x) g(x)| | x X}.

Definition 10.1. The family F C(X) is called equicontinuous if for every

> 0 there is > 0 such that

|f (x) f (y)| < for all x, y X satisfying d(x, y) < and all f F.

The family F is called equibounded if there is a constant M such that

|f (x)| M

for all f F and all x X.

Example 10.2. Let (X, d) be a metric space and let M > 0. By F denote the
family of functions f : X R satisfying

|f (x) f (y)| M d(x, y) for all x, y X.

Then the family F is equicontinuous. To see this, take > 0 and set = /M .
Then for every y B (x) and every f F, we have

|f (x) f (y)| M d(x, y) < M /M = .

Let M > 0 and let F be the set all differentiable functions f : [a, b] R
satisfying |f ! (x)| M for all x (a, b). In view of the mean value theorem,

|f (x) f (y)| M |x y|

for all x, y (a, b). Using the previous example, F is a equicontinuous family.
If (X, d) is a metric space, we call a subset F X precompact (or relatively
compact) if F is compact in X.
Theorem 10.3 (AscoliArzela Theorem). Let (X, d) be a compact space.
A subset F of C(X) is relatively compact if and only if F is equibounded and
equicontinuous.
Proof. Assume that F is relatively compact. This is means that F is compact.
We claim that F is equibounded and equicontinuous. Since F compact, it is
totally bounded. In particular, F is totally bounded. This implies that F is
equibounded. To see that F is equicontinuous, take > 0. Then, there are
f1 , . . . , fN C(X) such that

F B/3 (f1 ) . . . B/3 (fN ). (1)

60
Each fi is uniformly continuous since (X, d) is compact. Hence there exists
> 0 such that
|fi (x) fi (y)| < /3 for all x, y such that d(x, y) < and all 1 i N .
Given f F, in view of (10) there is 1 j N such that
(f, fj ) < /3.
Now if x, y X satisfy d(x, y) < , then
|f (x) f (y)| |f (x) fj (x)| + |fj (x) fj (y)| + |fj (y) f (y)|
(f, fj ) |fj (x) fj (y)| + (f, jj ) < 3/3 =
showing that F is equicontinuous.
Conversely, assume that F is equibounded and equicontinuous. It suffices
to show that F is totally bounded. Indeed, if F is totally bounded, then F is
totally bounded and since C(X) is complete, the set F is also complete. Hence
F is compact. Take > 0. Since F is equicontinuous, for every x X there
exists x > 0 such that
|f (x) f (y)| < /4 for all y such that d(x, y) < x and all f F.
The collection {Bx (x)}xX is an open cover of a compact metric space X.
Hence there are x1 , . . . , xN such that
X = Bx1 . . . BxN .
In particular,
|f (x) f (xi )| < /4 for all x Bi (xi ) and all f F,
where we have abbreviated i = xi . Since F is equibounded, the set F :=
{f (xi )|1 i N, f F } is bounded. Since a bounded set in R (with the
standard metric) is totally bounded, there are points y1 , . . . , yK in R such that
!
F B/4 (yi ).
1iK

For any map : {1, . . . , N } {1, . . . , K}, define

F := {f F|f (xi ) B/4 (y(i) ), i = 1, . . . , N }.
Note that there are only finitely many sets F and that every f F belongs
to one of the sets F . We claim that the diameter of F is finite. Indeed, take
f, g F and x X. Then x Bi (xi ) for some i and
" "
|f (x) g(x)| |f (x) f (xi )| + "f (xi ) y(i) "
" "
+ "y(i) g(xi )" + |g(xi ) g(x)| 4 /4 = .
Hence (f, g) < showing that diam F . Consequently, F can be covered
by finitely many sets of diameter less than. Hence F is totally bounded and
the proof is complete. !

61
 A simple consequence we have the following corollary.
Corollary 10.4. Let X be a compact metric space and let (fn ) C(X) be
a sequence which is equibounded and equicontinuous in C(X). Then every the
sequence (fn ) has a uniformly convergent subsequence.

11 Structure of complete metric spaces-Baires

theorem
Let (X, d) be a metric space. A subset U of X is called dense if U = X. If U
and V are open and dense, then U V is also open and dense. To see that U V
is dense, we have to show that O U V is non-empty for any open set O. Since
U is dense, there is u O U , and since O U is open, B(u, r) O U for some
r > 0. Since V is dense, B(u, r)V )= so that, )= B(u, r)V OU V . If U
and V are assumed to be dense but not necessarily open, then the intersection
U V does not have to be dense. For example, let U be the set of rational
numbers and V the set of irrational numbers Qc . Then both sets are dense in R
with the usual metric, however, U V = . Consider,
# now a sequence of dense
and open sets Un . In general, the intersection n1 Un may be empty.
Example 11.1. Consider (Q, d) with the usual metric d. Let {qn |n N} be
an enumeration of rational numbers, and let Un = Q \ {qn }. Then each Un
is open since#it is$ a complement
# % of&a closed set {qn }, and is dense . However,
U
n1 n = n1 Q \ {qn } = Q \ n1 {qn } = .

A subset F of X is called nowhere dense if (F ) = .

Theorem 11.2 (Baire). Let (X, d) be a complete metric space. Then:

#
(a) If {U n} is a sequence of open and dense subsets of X, then n1 Un is
dense.
&
(b) If {Fn } is a sequence of nowhere dense subsets of X, then Fn has empty
interior.
#
Proof. (a) It suffices to show that B(x, r) contains a point belonging to n1 Un
for any open ball B(x, r). Since U1 is open and dense, B(x, r) U1 is non-
empty and open. So, there exists an open ball B(x1 , R) with R < 1 such that
B(x1 , R) B(x, r) and B(x1 , R) U1 . Taking r1 < R, we get that B(x1 , r1 )
B(x, r) and B(x1 , r1 ) U1 . Similarly, since U2 is open and dense, there exists
x2 and r2 < 1/2 such that B(x2 , r2 ) B(x1 , r1 ) U2 . Continuing in this
way we find a sequence of balls B(xn , rn ) with rn < 1/n and B(xn+1 , rn+1 )
B(xn , rn ) Un . We claim that {xn } is Cauchy. By construction, Bn (xn , rn )
B(xk , rk ) for all n k. Given > 0 choose k N so that 1/k < /2. Then, if
n, m k,

d(xn , xm ) d(xn , xk ) + d(xk , xm ) < 1/k + 1/k < .

62
Because (X, d) is complete, {xn } converges, say to y. The point y lies in all
balls B(xk , rk ) since xn B(xk , rk ) for all n k and B(xk , rk ) is closed for all
k, so that after taking a limit as n , y B(xk , rk ) for all k. In particular,
y B(x1 , r1 ) # B(x, r) and y B(xn+1 , rn+1 ) Un for all n. Consequently,
y B(x, r) n1 Un , and the proof is finished. &
(b) Arguing& by contradiction assume that Fn has non-empty interior. So
B(x, r) Fn for some x and r > 0. Define Un = X \ Fn . Clearly, Un is
open and we claim that it is dense. Indeed, if for some open set V we have
V Un = #, then V X \ Un = Fn contradicting # that Fn has empty interior.
By (a), n1 Un is dense. So B(x, r) n1 Un )= . On the other hand,
& & $ & %
B(x, r) F Fn so that = B(x, r) X \ n1 Fn = B(x, r)
# $ %n #
n1 X \ Fn = B(x, r) n1 Un , a contradiction. !
Example 11.3. The metric space R with the standard metric space cannot
be written as a countable union of nowhere dense sets since it is complete. By
contrast, Q with the standard metric can be written as the union of one point
sets {qn }, where {qn | n N} is an enumeration of Q. Every one point set
{qn } is closed in Q and its interior is empty, so nowhere dense. This does not
contradict Baires theorem since Q with the standard metric is not complete.

Application of the Baires theorem

Theorem 11.4. There exists a continuous function f : [0, 1] R which is not
differentiable at any point x [0, 1).
Proof. Recall that f has a right-hand derivative at x if

lim [(f (x + h) f (x))/h] exists.

h0+

!
We denote this limit by f+ (x). In particular, if f is differentiable at x [0, 1)
then f+ (x) exists and is equal to f ! (x). Let
!

!
D = {f C([0, 1], R)| there exists x [0, 1) such that f+ (x) exists}

and let Dn,m be the set of all f C([0, 1], R) for which there exists some
x [0, 1 1/m] such that

|f (x + h) f (x)| n h for all h [0, 1/m].

&
We shall show that & D n,m Dn,m and that each Dn,m is nowhere dense. By
Theorem 11.2 (b) , n,m Dn,m has empty interior. Consequently, the set D has
empty interior showing that D )= C[0, 1].
&
Claim 1: D n,m Dn,m . To see this, let f D. Then there exists x [0, 1)
!
such that f+ (x) exists. In particular,
" "
" f (x + h) f (x) " " ! "
lim "" " = "f+ (x)" . (1)
h0+ h "

63
 " ! "
Take an integer n such that "f+ (x)" < n. Then there exists > 0 such that

|f (x + h) f (x)| n h for all 0 h .

Now choose m N large so that x 1 1/m and 1/m < . Then f Dn,m as
claimed.
Claim 2: Dn,m is closed. Take a sequence (fk ) Dn,m such that (fk , f ) 0
for some f C[0, 1]. To prove the claim we have to show that f Dn,m . Since
fk Dn,m , there is a sequence (xk ) satisfying 0 xk 1 1/m and

|fk (xk + h) fk (xk )| nh for all 0 h 1 1/m. (2)

Without loss of generality we may assume that xk x [0, 1 1/m]. Then,

using the triangle inequality,

|f (x + h) f (x )| |f (x + h) f (xk + h)| + |f (xk + h) fk (xk + h)|

+ |fk (xk + h) fk (xk )| + |fk (xk ) fk (x )|
+ |fk (x ) f (x )|
|f (x + h) f (xk + h)| + |fk (xk ) fk (x )|
+ 2d(fk , f ) + n h

for all 0 h 1/m. Since d(fk , f ) 0, |f (x + h) f (xk + h)| 0, and

|fk (xk ) f (xk )| 0, we conclude that

|f (x + h) f (x)| n h

for all 0 h 1/m. Consequently, f Dn,m showing that Dn,m is closed.

Claim 3: Dn,m = . This together with Claim 2 implies that Mm is nowhere
dense. To prove the claim it suffices to show that if f Dn,m , then any open ball
B contains g which doesnt belong to Dn,m . Take a piecewise linear function
"g !: [0,"1] R such that (f, g) = sup{|f (x) g(x)| | 0 x 1}
< and
"g+ (x)" > n for all x [0, 1). Then g B (f ) and g ) Dn,m . So, Dn,m = , as
claimed. !

64