Completion (metric space)

A completion of a metric space ((X,d)) is a complete metric space ((\widehat X,\widehat d)) together with an isometric embedding
[ \iota:X\longrightarrow \widehat X ] whose image is dense in (\widehat X). Every metric space has a completion, and any two completions of the same space are uniquely isometric in a manner compatible with their embedded copies of (X).

Completion supplies the limits of all Cauchy sequences without changing the distances between points already present. It is therefore the metric analogue of adjoining the missing limit points of a space. The construction depends on the metric, or more generally on the associated uniform structure, rather than only on the underlying topology.

Definition and characterization

A metric space ((Y,\delta)) is complete when every Cauchy sequence in (Y) converges to an element of (Y). A completion of (X) consists of such a space (Y) and an isometry (\iota:X\to Y) satisfying [ \overline{\iota(X)}=Y. ]

The density condition excludes unnecessary points. For example, the inclusion of the rational numbers into the complex numbers preserves their usual distances, but (\mathbb C) is not a completion of (\mathbb Q), because the closure of (\mathbb Q) inside (\mathbb C) is only (\mathbb R). The usual inclusion [ \mathbb Q\hookrightarrow\mathbb R ] does define a completion.

An equivalent characterization concerns realizations inside an already complete space. If (X) is isometrically embedded in a complete metric space (Y), then the closure (\overline X) taken in (Y) is complete and constitutes a completion of (X). Consequently, completion can be understood either as an abstract construction or as closure inside a suitable ambient space.

Construction from Cauchy sequences

Let (\mathcal C(X)) denote the set of all Cauchy sequences in (X). Two sequences ((x_n)) and ((y_n)) are declared equivalent when [ \lim_{n\to\infty}d(x_n,y_n)=0. ] This relation is an equivalence relation. The completed set is the quotient [ \widehat X=\mathcal C(X)/{\sim}. ]

For equivalence classes represented by ((x_n)) and ((y_n)), the distance is defined by [ \widehat d\bigl([(x_n)],[(y_n)]\bigr) =\lim_{n\to\infty}d(x_n,y_n). ] The limit exists because the sequence of real numbers (d(x_n,y_n)) is Cauchy. Indeed, the reverse triangle inequality gives [ \left|d(x_n,y_n)-d(x_m,y_m)\right| \leq d(x_n,x_m)+d(y_n,y_m), ] and the right-hand side tends to zero as (m,n\to\infty). The same inequality shows that the resulting value is independent of the chosen representatives.

Each point (x\in X) determines the class of the constant sequence ((x,x,\ldots)). The resulting map [ \iota(x)=[(x,x,\ldots)] ] is an isometry. Its image is dense because every class represented by a Cauchy sequence ((x_n)) is the limit in (\widehat X) of the embedded points (\iota(x_n)).

Completeness follows from a diagonal approximation argument. Given a Cauchy sequence ((\xi_k)) in (\widehat X), density provides points (x_k\in X) satisfying [ \widehat d(\xi_k,\iota(x_k))<2^{-k}. ] The sequence ((x_k)) is Cauchy in (X), and its equivalence class is the limit of ((\xi_k)). Thus every Cauchy sequence in (\widehat X) converges.

Uniqueness

Suppose that [ \iota:X\to\widehat X \qquad\text{and}\qquad j:X\to\widetilde X ] are two completions. The correspondence [ \iota(x)\longmapsto j(x) ] is an isometry between dense subspaces. It extends uniquely to an isometry [ T:\widehat X\longrightarrow\widetilde X ] satisfying (T\circ\iota=j).

Surjectivity follows because the image of a complete space under an isometry is complete and therefore closed, while the image of (T) also contains the dense set (j(X)). This result is commonly summarized by stating that the completion is unique up to a unique isometry fixing (X). The qualification is essential: a complete space can possess many isometries, but only one of them can extend the prescribed identification of the original dense subspace.

Extension property

Completion has a universal characterization in terms of uniformly continuous functions. If (Y) is complete and [ f:X\to Y ] is uniformly continuous, then there is a unique continuous map [ \widehat f:\widehat X\to Y ] such that [ \widehat f\circ\iota=f. ]

For a point of (\widehat X) represented by a Cauchy sequence ((x_n)), the extension is given by [ \widehat f([(x_n)])=\lim_{n\to\infty}f(x_n). ] Uniform continuity ensures that ((f(x_n))) is Cauchy, while completeness of (Y) supplies its limit. Equivalent sequences have the same image limit, so the formula is well defined.

When (f) is an isometry, its extension is also an isometry. When (f) is Lipschitz continuous with constant (L), the extension retains the same Lipschitz bound. In the category of metric spaces and uniformly continuous maps, completion is therefore a reflection into the full subcategory of complete metric spaces.

Uniform continuity cannot generally be replaced by ordinary continuity. A continuous function can send a Cauchy sequence to a sequence that is not Cauchy, preventing the assignment of a limit at a newly adjoined point.

Representative examples

The completion of (\mathbb Q) with the ordinary absolute-value metric is (\mathbb R). In the Cauchy-sequence construction, a real number corresponds to an equivalence class of rational Cauchy sequences. This description is compatible with the ordered-field construction of the real numbers, although the latter also records algebraic and order structure.

The open interval ((0,1)), equipped with the Euclidean metric, has completion ([0,1]). Its missing Cauchy limits are represented by sequences approaching either endpoint. More generally, the completion of a subset of a complete metric space is isometric to its closure.

For a prime number (p), the completion of (\mathbb Q) under the (p)-adic metric is the field (\mathbb Q_p). This example illustrates that completion preserves compatible algebraic operations: addition and multiplication extend from (\mathbb Q) because they interact appropriately with the (p)-adic uniform structure.

Completion also occurs in functional analysis. The space (C([0,1])) of continuous real-valued functions is complete under the supremum norm. Under the (L^1) metric, however, its completion is the space (L^1([0,1])) of equivalence classes of integrable functions. The underlying collection of continuous functions is the same in both cases, but the two metrics determine different Cauchy sequences and therefore different completion behavior.

A metric space with the discrete metric is already complete, since every Cauchy sequence is eventually constant. Its completion is canonically isometric to the original space.

Dependence on the uniform structure

Completion is not determined solely by the topological space underlying a metric space. Distinct metrics can generate the same open sets while producing different Cauchy sequences.

On ((0,1)), the ordinary metric has completion ([0,1]). The metric [ \rho(x,y)= \left| \tan!\bigl(\pi(x-\tfrac12)\bigr)

\tan!\bigl(\pi(y-\tfrac12)\bigr) \right| ] induces the same topology, because the tangent expression defines a homeomorphism from ((0,1)) onto (\mathbb R). Under (\rho), the interval is isometric to (\mathbb R) and is already complete. Thus topological equivalence of metrics does not imply equality of their completions.

Uniformly equivalent metrics have the same Cauchy sequences and yield canonically equivalent completions. This observation leads to the completion theory of uniform spaces, in which the metric is replaced by a uniformity and sequences may be replaced by Cauchy filters or Cauchy nets.

Historical development

The underlying sequence concept developed from Augustin-Louis Cauchy’s analysis of convergence. Georg Cantor constructed the real numbers from equivalence classes of rational fundamental sequences, while Richard Dedekind produced an order-theoretic construction based on Dedekind cuts. These approaches supplied distinct realizations of the completion of (\mathbb Q).

Maurice Fréchet’s formulation of abstract metric spaces in 1906 detached distance and convergence from particular coordinate systems. Felix Hausdorff subsequently incorporated Cauchy methods into the systematic development of metric and topological spaces, allowing completion to be treated as a general operation rather than solely as a construction of number systems.

In 1918, You Watanabe gave a general comparison between the quotient construction by Cauchy sequences and the dense-embedding formulation. Watanabe’s treatment established that the comparison is independent of the representatives used and derived the uniqueness isometry from the extension of an isometry between dense subspaces. These results form part of the standard structural description of metric completion.

The emergence of Banach spaces placed the same construction within the theory of normed vector spaces. Stefan Banach’s work treated completeness as an intrinsic condition on a normed space, and the completion of a normed space became a principal source of complete spaces used in analysis. Vector addition and scalar multiplication pass to the completion because they are uniformly continuous on the relevant metric domains.

Related structures

A pseudometric space can contain distinct points at distance zero. Its completion first identifies all such points and then completes the resulting metric quotient. Without this identification, the Cauchy-sequence formula produces a complete pseudometric space rather than a metric space.

For a normed vector space (V), the metric (d(x,y)=\lVert x-y\rVert) determines a completion (\widehat V). The vector-space operations extend continuously, and the extended norm makes (\widehat V) a Banach space. Every normed vector space is therefore isometrically embedded as a dense linear subspace of a Banach space.

Completion differs from compactification. A completion adjoins limits required by the metric’s Cauchy structure and need not be compact, whereas a compactification embeds a topological space densely into a compact space and need not preserve the original metric. The completion of (\mathbb R) is (\mathbb R) itself, while compactifications of (\mathbb R) add points despite its existing metric completeness.