A Cauchy sequence can describe a point that a metric space has failed to include. The sequence is Cauchy in but has no limit there. Rather than repair each missing limit separately, completion adds all of them at once, while keeping the original distances. The construction gives a useful sufficient condition for extending maps: uniform continuity guarantees an extension to any complete target. Mere continuity is not sufficient, although uniform continuity is not necessary for every particular extension.
Cauchy sequences as candidate points
Definition 1.2.1
(Equivalent Cauchy sequences and the completion)
Let be a nonempty metric space. Two Cauchy sequences and are equivalent if . This is an equivalence relation: the triangle inequality proves transitivity. Let be the set of equivalence classes and write for a class. Define
The limit exists because
so these real distances form a Cauchy sequence. If equivalent representatives are substituted, the distance changes by at most the two corresponding pointwise errors, both tending to zero. Thus the definition does not depend on representatives. Symmetry and the triangle inequality pass to limits; distance zero means exactly that the sequences are equivalent. Hence is a metric.
The map sends to the class of the constant sequence . It is an isometric embedding: . We may regard as a subset of the completion through , but an original point and an equivalence class are not literally the same object before this identification.
Density and completeness
Theorem 1.2.2
(Density and completeness of the completion)
is dense in , and is complete.
Proof
(Approximating completed points by original points)
For ,
The convergence follows from the Cauchy property of . Thus every completed point is a limit of embedded original points.
Now let be Cauchy in . By density choose with . The triangle inequality shows that is Cauchy in : the first and third terms in
are small for large indices, and so is the middle term. Set . The density argument already proved gives , and then gives . No completeness of the original space was assumed.
Proof X-ray
The two uses of Cauchy sequences
An individual Cauchy sequence in defines a point of the completion, possibly one already embedded from : if it converges to , its class is . Different sequences can define the same point. A class represents a genuinely added point precisely when its sequences do not converge in . To prove the new space complete, we start instead with a Cauchy sequence of these points, approximate its th term by an original point within , and show those original points form a Cauchy sequence. The second construction is the step that prevents the completion from requiring further points.
Worked example 1.2.3
(The rationals complete to the reals)
By the standard density of in , the rationals with ordinary distance embed densely and isometrically into . Completeness of implies every rational Cauchy sequence has a real limit, and two such sequences give the same limit exactly when their difference tends to zero. The uniqueness theorem proved below therefore identifies their completion with up to isometry. This argument uses the already established completeness of the real numbers; it is not a circular construction of the reals from rationals.
Extending a uniformly continuous map
Theorem 1.2.4
(Extension of uniformly continuous maps)
Let be a complete metric space and uniformly continuous. There is a unique uniformly continuous satisfying for every .
Proof
(Definition by limits and its uniform continuity)
If is Cauchy in , uniform continuity makes Cauchy in . Completeness supplies its limit. Define
If and are equivalent, uniform continuity gives , so the definition is independent of representative. It agrees with on constant sequences.
To check uniform continuity of the extension, fix and choose such that implies . If , choose so that are within of and are within of respectively. Such choices exist by the defining limits. Then , so the triangle inequality bounds by .
Finally, any continuous extension must agree with this one on limits of points of the dense set . This proves uniqueness even among continuous extensions, provided an extension exists.
Worked example 1.2.5
(Two bounded functions on )
Let and on , with complete target ; the completion of is . The function is uniformly continuous. Setting and gives a continuous function on the compact interval , because ; a continuous function on a compact metric space is uniformly continuous, and so is its restriction to . The extension theorem then gives for every sequence in , again because .
For , take and . Both sequences are Cauchy and , so they represent the same point of the completion. But and for every . The recipe would assign two different values to one point, so has no continuous extension to , although it is bounded and continuous. A bounded image does not replace uniform continuity: what fails is independence of the representative.
Common mistake
(The extension theorem is not a characterization)
Uniform continuity is sufficient, not necessary in every instance. For example, on the already complete space extends continuously to its completion as itself, although it is not uniformly continuous on . Completeness of is essential: the identity from to is uniformly continuous, but cannot extend continuously from to , since rational sequences converging to would need a rational limit. The target must contain the required limits.
The completion is unique up to isometry
Theorem 1.2.6
(Uniqueness of the completion)
Let be a complete metric space and an isometric embedding with dense image. Then there is exactly one continuous map with , and it is an isometry of onto .
Proof
(Extending the second embedding)
Suppose another complete metric space contains an isometric copy densely. Apply the extension theorem to . For two classes represented by , continuity of distance gives
Thus is an isometry. Its image is complete because is complete, and a complete subset of a metric space is closed: an ambient limit of a sequence in it is also the limit supplied by completeness. The image contains the dense set , so it is all of . The surjective isometry is uniquely determined by requiring it to agree with on . Completion is therefore unique in this precise sense, not unique as a literal set of objects.
Concept lensStructural
The completion as a closure inside a function space
The uniqueness theorem lets any dense isometric copy in a complete space serve as the completion, so Cauchy classes are one model among several. Fix and send to in the complete space of bounded continuous real functions on with the supremum norm. Then , and , attained at . The closure of is closed in a complete space, hence complete, and contains the copy of densely: it is up to isometry.
Topology alone does not determine completeness
Worked example 1.2.7
(An incomplete metric with the usual topology)
On define . The map is an isometry from onto , so this space has completion . The sequence is Cauchy for and has no -limit in ; its new limit corresponds to .
Common mistake
(Completeness is not a topological property)
The metric induces the ordinary topology: is a homeomorphism onto its open interval. Ordinary Euclidean distance on is complete, while is not. Completeness of a chosen metric is not determined by its topology. The separate property of whether a topology admits some compatible complete metric is a different question. The two points added by the completion correspond to escape toward positive and negative infinity.
Counterexample mode
Continuity is not enough for extension
The function is continuous on , with complete target , but has no continuous extension to its completion : along the values have no finite limit. This does not contradict the extension theorem, because is not uniformly continuous. For example, while .
Practice with solutions
Quick check
(Show that every Cauchy sequence in a metric space is bounded.)
Solution · Answer
Choose so that for . The finite initial segment has finite maximum distance from , so one ball contains the whole sequence.
Quick check
(If is already complete, prove that its embedding is onto.)
Solution · Answer
Every class has a representative converging to some . It is equivalent to the constant sequence at , so the class is .
Quick check
(Does on extend uniformly continuously to ? Give the extension and justify uniform continuity.)
Solution · Answer
Yes: on . The extension is on .
Quick check
(In the metric above, identify the limit in the completion of the sequence and explain why it is distinct from the limit of .)
Solution · Answer
It corresponds to . The two completed points have distance , so they are distinct.