Complete Metric Spaces

Denition 1. Let (X, d) be a metric space. A sequence (xn ) in X is called a Cauchy sequence if for any > 0, there is an n N such that d(xm , xn ) < for any m n , n n . Theorem 2. Any convergent sequence in a metric space is a Cauchy sequence. Proof. Assume that (xn ) is a sequence which converges to x. Let > 0 be for all n N . Let m, given. Then there is an N N such that d(xn , x) < 2 n N be such that m N , n N . Then d(xm , xn ) d(xm , x) + d(xn , x) < Hence (xn ) is a Cauchy sequence. Then converse of this theorem is not true. For example, let X = (0, 1]. 1 Then ( n ) is a Cauchy sequence which is not convergent in X . Denition 3. A metric space (X, d) is said to be complete if every Cauchy sequence in X converges (to a point in X ). Theorem 4. A closed subset of a complete metric space is a complete subspace. Proof. Let S be a closed subspace of a complete metric space X . Let (xn ) be a Cauchy sequence in S . Then (xn ) is a Cauchy sequence in X and hence it must converge to a point x in X . But then x S = S . Thus S is complete. Theorem 5. A complete subspace of a metric space is a closed subset. Proof. Let S be a complete subspace of a metric space X . Let x S . Then there is a sequence (xn ) in S which converges to x (in X ). Hence (xn ) is a Cauchy sequence in S . Since S is complete, (xn ) must converge to some point, say, y in S . By the uniqueness of limit, we must have x = y S . Hence S = S , i.e. S is closed. Denition 6. Let A be a nonempty subset of a metric space (X, d). The diameter of A is dened to be diam(A) = sup{d(x, y ) | x, y A}. We say that A is bounded if diam(A) is nite. Theorem 7. Let (X, d) be a complete metric space. If (Fn ) is a sequence of nonempty closed subsets of X such that Fn+1 Fn for all n N and (diam(Fn )) converges to 0, then n=1 Fn is a singleton. 1 + = . 2 2

Proof. Assume that (X, d) is a complete metric space. Let (Fn ) be a sequence of nonempty closed subsets of X such that Fn+1 Fn for every n N and (diam(Fn )) converges to 0. First, we show that n=1 Fn is nonempty. For each n N, choose xn Fn . To show that (xn ) is a Cauchy sequence, let > 0. Since (diam(Fn )) converges to 0, there is an N N such that diam(FN ) < . Let m, n N with n N , m N . Then xm Fm FN , xn Fn FN . Hence d(xm , xn ) diam(FN ) < . Thus (xn ) is a Cauchy sequence in X . Since X is complete, it follows that (xn ) is a convergent sequence in X . Let x = lim xn .
n

We now show that x n=1 Fn . Let n N. Note that xm Fm Fn for any m n. Thus the sequence (xn , xn+1 , . . . ) is a sequence in Fn and is a n = Fn . subsequence of (xn ), so it converges to x. This implies that x F Thus x n=1 Fn is a singleton. To see this, let n=1 Fn . Next, we show that x, y n=1 Fn and > 0. Then there is an N N such that diam(FN ) < . Since x, y FN , it follows that d(x, y ) diam(FN ) < . This is true for any > 0. Hence d(x, y ) = 0, which means x = y . Denition 8. Let f be a function from a metric space (X, d) into a metric space (Y, ). We say that f is uniformly continuous if given any > 0, there exists a > 0 such that for any x, y X , d(x, y ) < implies (f (x), f (y )) < . Theorem 9. A uniformly continuous function maps Cauchy sequences into Cauchy sequences. Proof. Let f : (X, d) (Y, ) be a uniformly continuous function. Let (xn ) be a Cauchy sequence in X . To see that (f (xn )) is a Cauchy sequence, let > 0. Then there is a > 0 such that x, y X, d(x, y ) < (f (x), f (y )) < . Thus there exists an N N such that d(xm , xn ) < for any m, n N . It follows that (f (xm ), f (xn )) < for any m, n N . Hence (f (xn )) is a Cauchy sequence in Y . Remark. If f is not uniformly continuous, then the theorem may not be true. 1 1 is continuous on (0, ) and (xn ) = ( n ) is a Cauchy For example, f (x) = x sequence in (0, ) but (f (xn )) = (n) is not a Cauchy sequence.

Denition 10. Let f be a function from a metric space (X, d) into a metric space (Y, ). We say that f is an isometry if d(a, b) = (f (a), f (b)) for any a, b X. Theorem 11. Let f : (X, d) (Y, ) be an isometry. Then it is injective and uniformly continuous. Moreover, its inverse f 1 : (f [X ], ) (X, d) is also an isometry. Proof. Let f : (X, d) (Y, ) be an isometry. Let > 0. Choose = > 0. Let a, b X be such that d(a, b) < . Then (f (a), f (b)) = d(a, b) < = . Hence f is uniformly continuous. Next, let a, b X be such that f (a) = f (b). Thus d(a, b) = (f (a), f (b)) = 0. This shows that a = b. Hence f is injective. To see that f 1 is an isometry, let y , z f [X ] and let a, b X be such that f (a) = y and f (b) = z . Thus (y, z ) = (f (a), f (b)) = d(a, b) = d(f 1 (y ), f 1 (z )). Hence f 1 is an isometry. Theorem 12. Let A be a dense subset of a metric space (X, d). Let f be a uniformly continuous function (isometry) from A into a complete metric space (Y, ). Then there is a unique uniformly continuous function (isometry) g from X into Y which extends f . Proof. We will give a proof only for a uniformly continuous function. The proof for an isometry is similar and somewhat easier. Let (X, d) be a metric space and (Y, ) a complete metric space. Let A be a dense subset of X and let f be a uniformly continuous from A into Y . Step 1 : dene a function g : X Y . For each x X = A, there is a sequence (xn ) in A which converges to x. Then (xn ) is a Cauchy sequence in X . Thus (f (xn )) is a Cauchy sequence in Y . Since Y is complete, (f (xn )) is a convergent sequence. Dene g (x) = lim f (xn )
n

for any x X , where (xn ) is a sequence in A which converges to x. Step 2 : g is well-dened, i.e., independent of the choice of (xn ). Let (xn ) and (yn ) be any sequence in A which converges to x A = X . Then the sequence (x1 , y1 , x2 , y2 , . . . , xn , yn , . . . ) must converge to x. Hence, the sequence (f (x1 ), f (y1 ), f (x2 ), f (y2 ), . . . , f (xn ), f (yn ), . . . ) converges 3

to some point z Y . Since (f (x1 ), f (x2 ), . . . ) and (f (y1 ), f (y2 ), . . . ) are its subsequences, they must also converge to z . Hence z = g (x) does not depend on the choice of the sequences. Step 3 : g is an extension of f . Let a A and let an = a for each n N. Then (an ) is a sequence in A which converges to a. Hence g (a) = limn f (an ) = f (a). This shows that g is an extension of f . Step 4 : g is uniformly continuous on X . Let > 0. Then there is a > 0 such that a, b A, d(a, b) < (f (a), f (b)) < . 3 Let x, y X be such that d(x, y ) < . Then there are sequence (xn ) and (yn ) in A such that xn x and yn y . Hence f (xn ) g (x) and f (yn ) g (y ). Choose N N such that d(x, y ) d(x, y ) and d(yN , y ) < 2 2 and (f (xN ), g (x)) < and (f (yN ), g (y )) < . 3 3 d(xN , x) < Then d(xN , yN ) d(xN , x) + d(x, y ) + d(y, yN ) < , which implies that (f (xN ), f (yN )) < , by the uniform continuity of f on A. Hence, (g (x), g (y )) (g (x), f (xN )) + (f (xN ), f (yN )) + (f (yN ), g (y )) < + + = . 3 3 3 This shows that g is uniformly continuous on X . Step 5 : g is unique. Let g and h be (uniformly) continuous functions on X which extends f on a dense subset A. To see that g = h, let x X . Then there is a sequence (xn ) in A which converges to x. By continuity of g and h, g (x) = lim g (xn ) = lim h(xn ) = h(x).
n n

(0.1) (0.2)

Hence g = h. Denition 13. A completion of a metric space (X, d) is a pair consisting of a complete metric space (X , d ) and an isometry of X into X such that [X ] is dense in X . 4

Theorem 14. Every metric space has a completion. Proof. Let (X, d) be a metric space. Denote by C [X ] the collection of all Cauchy sequences in X . Dene a relation on C [X ] by (xn ) (yn ) lim d(xn , yn ) = 0.
n

It is easy to see that this is an equivalence relation on C [X ]. Let X be the set of all equivalence classes for : X = { [(xn )] : (xn ) C [X ] }. Dene d : X X [0, ) by d ([(xn )], [(yn )]) = lim d(xn , yn ),
n

where [(xn )], [(yn )] X . To show that d is well-dened, let (xn ) and (yn ) be two Cauchy sequences in X such that (xn ) (xn ) and (yn ) (yn ). Then
n

lim d(xn , xn ) = lim d(yn , yn ) = 0.

By the triangle inequality, d(xn , yn ) d(xn , xn ) + d(xn , yn ) + d(yn , yn ) and d(xn , yn ) d(xn , xn ) + d(xn , yn ) + d(yn , yn ). Hence, |d(xn , yn ) d(xn , yn )| d(xn , xn ) + d(yn , yn ) 0. Since both (d(xn , yn )) and (d(xn , yn )) are convergent, this shows that
n

lim d(xn , yn ) = lim d(xn , yn ).

Thus d is well-dened. Next, we show that d is a metric on X . Let [(xn )], [(yn )], [(zn )] X . Then d ([(xn )], [(yn )]) = 0 lim d(xn , yn ) = 0 (xn ) (yn ) [(xn )] = [(yn )].
n

Also, d ([(xn )], [(yn )]) = lim d(xn , yn ) = lim d(yn , xn ) = d ([(yn )], [(xn )]).
n n

Since d(xn , zn ) d(xn , yn ) + d(yn , zn ),

lim d(xn , zn ) lim d(xn , yn ) + lim d(yn , zn ).

Thus d ([(xn )], [(zn )]) d ([(xn )], [(yn )]) + d ([(yn )], [(zn )]). Hence d is a metric on X . For each x X , let x = [(x, x, . . . )] X , the equivalence classes of the constant sequence (x, x, . . . ). Dene : X X by (x) = x . Then for any x, y X , d ((x), (y )) = d ( x, y ) = lim d(x, y ) = d(x, y ).
n

Hence is an isometry from X into X . To show that [X ] is dense in X , let x = [(xn )] X and let > 0. Since (xn ) is a Cauchy sequence, there exists an N N such that for any m, n N , d(xm , xn ) < . Let z = xN . 2 Then z [X ] and d (x , z ) = lim d(xn , z ) = lim d(xn , xN )
n n

< . 2

Thus z Bd (x , ) [X ]. Hence, [X ] is dense in X . Finally we show that (X , d ) is complete. To establish this, we apply the following lemma of which proof is left as an exercise: Lemma 15. Let (X, d) be a metric space and A a dense subset such that every Cauchy sequence in A converges in X . Prove that X is complete. Hence, it suces to show that every Cauchy sequence in the dense subspace [X ] converges in X . Let (zk ) be a Cauchy sequence in [X ], where each zk is represented by the Cauchy sequence (zk , zk , . . . ). Since is an isometry, d(zn , zm ) = d (zn , zm ) for each m, n.

Hence, (z1 , z2 , z3 , . . . ) is a Cauchy sequence in X . Let z = [(z1 , z2 , z3 , . . . )] X . To show that (zk ) converges to z , let > 0. Then there is an N N such that d(zk , zn ) < for any k , n N . Hence, for each k N , 2 d (zk , z ) = lim d(zk , zn ) < . n 2 This shows that (zk ) converges to a point z in X and that X is complete. Before proving that a completion of a metric space is unique up to isometry, we will give an alternative denition of a completion in terms of a universal mapping property. We state this fact in the following denition: Theorem 16 (Universal Mapping Property). Let (X, d) be a metric space, (X , d ) a complete metric space and : (X, d) (X , d ) an isometry. Then [X ] is dense in X if and only if it satises the following universal mapping property: 6

Given any complete metric space (Y, ) and an isomtry f : X Y , there exists a unique isometry F : X Y such that F = f . X?

?? ?? ? f ??? 

/ X     F  

Proof. Assume that [X ] is dense in X . We will show that it satises the universal mapping property. Let (Y, ) be a complete metric space and f : X Y an isometry from X into Y . Since is an isometry, is 1-1. Thus 1 : [X ] X is an isometry from [X ] onto X . Since f is an isometry from X onto Y , it follows that h := f 1 is an isometry from a dense subset [X ] of X into a complete metric space Y . Hence it can be extended uniquely into an isometry F from X into Y . Then for any x X F (x) = h((x)) = f 1 ((x)) = f (x). Thus F = f . If G : X Y is another isometry such that G = f , then F = G on the dense subset [X ] of X ; hence they must be equal on X , i.e. F = G. Conversely, assume that it satises universal mapping property and show that [X ] is dense in X . Since [X ] is closed in a complete metric space X , [X ] is also complete. Let Y = [X ] and f = . By the universal mapping property, there is a unique isometry F : X Y such that F = f = . This shows that F is the identity on the subspace [X ]. It implies that F is the identity on [X ]. From this, we must have X = [X ]. Theorem 17. A completion of a metric space is unique up to isometry. More , d2 )} are two completions of (X, d), precisely, if {1 , (X1 , d1 )} and {2 , (X2 then there is a unique isometry f from X1 onto X2 such that f 1 = 2 .
Proof. First, letting Y = X2 and f = 2 , by the universal mapping property of (X1 , d1 ) there is a unique isometry F : X1 X2 such that F 1 = 2 . Similarly, letting Y = X1 and f = 1 , by the universal mapping property of (X2 , d2 ) there is a unique isometry G : X2 X1 such that G 2 = 1 . Hence, G F is an isometry on X1 such that G F 1 = 1 . But then is an isometry on X 1 = 1 . By the the identity map IX1 1 such that IX1 . By the same argument, we can uniqueness property, we have G F = IX1 . This shows that F and G are inverses of each other. show that F G = IX2 Thus F is an isometry from (X1 , d1 ) onto (X2 , d2 ).

Denition 18. A function f : (X, d) (X, d) is said to be a contraction map if there is a real number k < 1 such that d(f (x), f (y )) k d(x, y ) for all x, y X. 7

Theorem 19. Let f be a contraction map of a complete metric space (X, d) into itself. Then f has a unique xed point. Proof. Let (X, d) be a complete metric space and f : X X a d-contractive map. Then there is a real number k < 1 and d(f (x), f (y )) kd(x, y ) for all x, y in X . Fix x0 X and let xn = f (xn1 ) for each n N. Then d(x1 , x2 ) = d(f (x0 ), f (x1 )) k d(x0 , x1 ), d(x2 , x3 ) = d(f (x1 ), f (x2 )) k d(x1 , x2 ) k 2 d(x0 , x1 ). Assume that d(xm1 , xm ) k m1 d(x0 , x1 ). Then d(xm , xm+1 ) = d(f (xm1 ), f (xm )) k d(xm1 , xm ) k m d(x0 , x1 ). By induction, we have d(xm , xm+1 ) k m d(x0 , x1 ) for each m N. Hence, for any m, n N with m > n, d(xn , xm ) d(xn , xn+1 ) + d(xn+1 , xn+2 ) + + d(xm1 , xm ) k n d(x0 , x1 ) + k n+1 d(x0 , x1 ) + + k m1 d(x0 , x1 ) (k n + k n+1 + k n+2 + ) d(x0 , x1 ) kn d(x0 , x1 ). = 1k Since 0 k < 1, the sequence (k n ) converges to 0. This implies that (xn ) is a Cauchy sequence. Since X is complete, the sequence (xn ) is convergent to, say, x in X . To show that f (x) = x, let > 0. Then there is an N N for any n N . Hence d(f (xN ), x) = d(xN +1 , x) < 2 . such that d(xn , x) < 2 Therefore d(f (x), x) d(f (x), f (xN )) + d(f (xN ), x) k d(x, xN ) + d(xN +1 , x) < k + < + = . 2 2 2 2 Thus f (x) = x. Now we show that the xed point is unique. Let x and y be xed points of f . Then d(x, y ) = d(f (x), f (y )) k d(x, y ). If d(x, y ) > 0, then 1 k , a contradiction. Thus d(x, y ) = 0. This shows that x = y . Hence, f has a unique xed point. Theorem 20 (Baires Theorem). If X is a complete metric space, the intersection of a countable number of dense open subsets is dense in X . Proof. In the proof of this theorem, we will denote the closed ball centered at x with radius r by B [x, r]: B [x, r] = {y X | d(y, x) r}. 8

Note that any open set in a metric space contains a closed ball. Indeed, if we shrink the radius of an open ball slightly, we obtain a closed ball contained in that open ball. Suppose that V1 , V2 , . . . are dense and open in X and let W be a nonempty open set in X . We will show that

(
n=1

Vn ) W = .

Since V1 is dense in X , W V1 is a nonempty open set. Hence, we can nd x1 X and 0 < r1 < 1 such that B [x1 , r1 ] W V1 . (0.3)

If n 2 and xn1 and rn1 are chosen, the denseness of Vn shows that Vn B (xn1 , rn1 ) is a nonempty open set, and therefore we can nd xn X and 1 such that 0 < rn < n B [xn , rn ] Vn B (xn1 , rn1 ). (0.4)

By induction, this process produces the sequence {xn } in X . If m, n N , then xm and xn are in B (xN , rN ), and thus 2 . N Hence, {xn } is a Cauchy sequence. Since X is complete, xn converges to some x X . If k n, then xk lies in a closed set B [xn , rn ]. Thus, x B [xn , rn ] for all n 1. By (0.3), x W V1 , and by (0.4), we have x Vn for all n 2. Hence, d(xm , xn ) d(xm , xN ) + d(xN , xn ) < 2 rN <

x
n=1

Vn

W.

We can conclude that the intersection of all Vn is dense in X . Remark. The completeness assumption is necessary in this theorem as the following example illustrates. Let X = Q . Write Q = { rn | n N } and let Gn = Q {rn } for each n N. Then Gn is open and dense in Q for each n, but n=1 Gn = . Corollary 21. If a complete metric space is a union of countably many closed sets, then at least one of the closed sets has nonempty interior. Proof. Let X be a complete metric space. Assume that X = n=1 Fn , where c each Fn is closed. For each n N, let Gn = Fn . Then n=1 Gn = . By Baires theorem, there exists an open set Gn which is not dense in X . Thus, Gn = X . But then Int Fn = X Gn , and hence Fn has nonempty interior. 9