Lebesgue Covering Lemma

S. Kumaresan
Dept. of Math. & Stat.
University of Hyderabad
Hyderabad 500 046
[email protected]
Definition 1. Given an open cover {Ui : i I} of a metric space (X, d), we say that a
positive number is a Lebesgue number of the cover, if for any subset A X whose diameter
is less than , there exists i I such that A Ui .
Remark 2. If is a Lebesgue number of the cover and 0 < 0 , then 0 is also a Lebesgue
number of the given open cover.
In general, an open cover may not have a Lebesgue number.
Ex. 3. Let X = (0, 1) with the usual metric. Let Un := (1/n, 1). Then {Un : n N} is an
open cover of X. Does there exist a Lebesgue number for this cover?
Theorem 4 (Lebesgue Covering Lemma). Let (X, d) be a compact metric space. Let {Ui :
i I} be an open cover of X. Then a Lebesgue number exists for this cover.
We give three proofs of this result.
Proof 1. For x X, there is an i(x) such that x Ui(x) and an r(x) > 0 such that
B(x, 2r(x)) Ui(x) . (Why?) There exist finitely many xk , 1 k n such that X =
k B(xk , rk ) where rk := r(xk ). Let be any positive real such that < min{rk }. Let A be any
subset with diam (A) < . Let a A. Then a B(xk , rk ) for some k. Let x A be arbitrary.
Then d(x, xk ) d(x, a) + d(a, xk ) < + rk < 2rk . Thus, A B(xk , 2rk ) Ui(xk ) .
Proof 2. Suppose that the result is not true. Then, for any = 1/n, there is a subset An with
diameter less than 1/n and such that it is not a subset of Ui for any i. Choose any xn An .
Then the sequence (xn ) has a convergent subsequence (xnk ) such that xnk p in X. Let
p Ui . Let r > 0 be such that B(p, 2r) Ui . Choose k so large that xnk B(p, r) and
1/nk < r. Now if a Ank is any element, then, d(a, p) d(a, xnk ) + d(xnk , p) < 2r. That is,
Ank B(p, 2r) Ui , contradicting our assumption on the An s.
Proof 3. We may assume that the given cover is finite, say, {Ui }1in . Let fi (x) := d(x, X\Ui ).
Then fi are continuous and fi (x) = 0 iff x X \ Ui , i.e., iff x
/ Ui . Let f := max{fi }. Then
f is continuous on X and f (x) = 0 iff x
/ Ui for all i, which is not possible, as Ui s cover X.
Thus, f (x) > 0 for all x X. Let := inf{f (x) : x X}. Then > 0. (Why?) Let A be any
subset with diam (A) < . Let a A be arbitrary. Then f (a) and hence fi (a) for
some i. Hence a Ui . If x A is any point, then x Ui . For, otherwise, x X \ Ui so that
d(a, x) d(a, X \ Ui ) = fi (a) . Hence diam (A) , a contradiction. Hence A Ui .
1

Ex. 5. Let f : (X, d) (Y, d) be continuous. Assume that X is compact. Prove that f is
uniformly continuous (i) using the theorem and (ii) imitating the first (second) proof of the
theorem.
Theorem 7 gives an interesting converse.
Definition 6. We say that a metric space has the Lebesgue number property, if every open
cover of X has a Lebesgue number.
Theorem 7. The following are equivalent for a metric space X:
1. X has Lebesgue number property.
2. Every continuous map from X to another metric space is uniformly continuous.
3. Every real valued continuous function on X is uniformly continuous.
Proof. (1) = (2): Let f : X Y be a continuous map from X to another metric space
Y . Let > 0 be given. Since f is continuous at x X, there exists a x > 0 such that for
any x0 B(x, x ), we have d(f (x0 ), (x)) < /2. Now, the collection {B(x, 21 x ) : x X}
is an open cover of X. Let be a Lebesgue number of the cover. Let x1 , x2 X with
d(x1 , x2 ) < /2. Since the diameter of B(x1 , /2) 2, there exists x X such that
B(x1 , /2) B(x, 21 x )). Hence d(x1 , x2 ) < x . It follows that
d(f (x1 ), f (x2 )) d(f (x1 ), f (x)) + d(f (x), f (x2 )) < /2 + /2 = .
Thus f is uniformly continuous.
(2) = (3): Take Y = R in (2).
(3) = (1): We shall prove this contradiction. So, we assume that there exists an
open cover {U : I} of X which has no Lebesgue number. This means that given any
n N, we can find xn X such that B(xn , 1/n) is not contained in any of the U s. In
other words, given n N, there exists xn X such that B(xn , 1/n) \ U 6= emptyset for each
I. We claim that no B(xn , 1/n) is a singleton. For, otherwise, xn must be in some U .
Hence B(xn , 1/n) = {xn } U , contradicting our choice of xn . So, let yn B(xn , 1/n) with
yn 6= xn .
We claim that neither of the two sequences (xn ) and (yn ) can have a convergent subsequence. Assume the contrary. For instance, let us assume that (xnk ) is a convergent subsequence of (xn ) converging to some x X. If x U (which must happen for some I), then
there exists r > 0 such that B(x, r) U . Since xnk x as k , it follows that for some
k0 N, we have xnk B(x, r) for all k k0 . Since nk , we see that 1/nk < r d(x, xnk )
for all sufficiently large k. As a consequence, B(xnk , 1/nk ) B(x, r) U , a contradiction
to our choice of xn s. If (ynk ) is a convergent subsequence, converging to y X, clearly,
xnk y, impossible by what was seen just now.
We now construct two closed subsets A and B out of these two sequences inductively.
Let n1 = 1. We assume that x1 and y1 are already in A and B. We select n2 > n1 such
that xn2 6= xn1 and yn1 6= yn1 . This is possible, since otherwise, for all n > n1 , xn = x1 etc.
Hence (xn ) and (yn ) will have convergent subsequences, contradicting our claim in the last
paragraph. Assume that we have found n1 < < nk such that xn1 , . . . , xnk and xn1 , . . . , xnk
are disjoint. We select nk+1 > nk so that xnk+1
/ {xn1 , . . . , xnk }, ynk+1
/ {yn1 , . . . , ynk } and
2

the sets Ak := {xnj : 1 j k + 1} and Bk := {ynj : 1 j k + 1} are disjoint. That this

is possible is seen as earlier.
For, if this is not possible, then for all n > nk , xn must lie in the finite set Ak Bk .
It follows that xn must be one of the elements in this set for infinitely many values
of n. But then (xn ) will have a convergent subsequence.
We claim A := {xnk } and B := {ynk } are disjoint and closed. They are disjoint by
construction. The set A is closed, if x is the limit of a sequence in A, then (xn ) has a
convergent subsequence. Hence no x X \ A could be the limit of a sequence in A. Hence
A is closed. Similarly, B is closed. Let us define f (a) = 0 for all a A and f (b) = 1 for
all b B. Then f is continuous on the closed set A B. By Urysohns lemma, there exists
a continuous extension, call it g, to X. The function g cannot be uniformly continuous.
For, let > 0 work for = 1. The d(xnk , ynk ) < 1/nk < for all sufficiently large k but
|g(xnk ) g(ynk )| = 1. So, we conclude that g is continuous but not uniformly continuous.
This contradicts our hypothesis on X.
Remark 8. Let X be a complete metric space such that every real valued continuous function
is uniformly continuous. Is X compact? No, not necessarily. Look at R with discrete topology.