1 Overview of Cauchy sequences and completeness
1.1 Cauchy sequences in metric spaces
In a metric space \((X,d)\), a sequence \((x_n)\) is called Cauchy if the points eventually become arbitrarily close to each other: for every \(\varepsilon>0\) there exists \(N\) such that \(d(x_m,x_n)<\varepsilon\) whenever \(m,n\ge N\). Informally, the sequence “settles down” even if it is not known whether it converges to a point of \(X\).
Cauchy sequences depend only on the metric. However, they also reflect the underlying notion of “approaching a limit,” which is formalized differently in other settings such as uniform spaces.
1.2 Completeness and convergence
A metric space \((X,d)\) is complete if every Cauchy sequence in \(X\) converges to a point of \(X\). Convergence means: there exists \(x\in X\) such that \(d(x_n,x)\to 0\).
Completeness is stronger than “no Cauchy sequence diverges wildly”; it guarantees that the limiting behavior already has a home in the space. If \(X\) is not complete, Cauchy sequences may represent “would-be limits” that lie outside \(X\).
1.3 Examples illustrating incompleteness
| A standard example is the rational numbers \(\mathbb{Q}\) with the usual metric \(d(x,y)= | x-y | \). Sequences of rationals can converge to real numbers that are irrational; such sequences are Cauchy in \(\mathbb{Q}\) but do not converge within \(\mathbb{Q}\). |
|---|
More generally, many spaces encountered in analysis and geometry are constructed first in a “non-complete” form and later enlarged by completion to ensure that limiting operations behave robustly.
2 Constructing the Cauchy completion
2.1 Equivalence relation on Cauchy sequences
2.1.1 Defining when two Cauchy sequences represent the same point
Let \((X,d)\) be a metric space. Consider the set \(\mathcal{C}\) of all Cauchy sequences in \(X\). Define an equivalence relation \(\sim\) on \(\mathcal{C}\) by declaring that two Cauchy sequences \((x_n)\) and \((y_n)\) are equivalent if \[ d(x_n,y_n)\to 0. \] Equivalently, for every \(\varepsilon>0\), there exists \(N\) such that \(d(x_n,y_n)<\varepsilon\) for all \(n\ge N\). Intuitively, the sequences approach the same “limit behavior,” even if that limit is not present in \(X\).
This relation is compatible with the Cauchy property: if two sequences are equivalent in this sense, then any sequence of “eventual neighborhoods” matches up in the limit.
2.2 Defining the metric on equivalence classes
2.2.1 Well-definedness of the distance function
Let \(\widehat{X}\) be the set of equivalence classes of Cauchy sequences. For classes represented by \((x_n)\) and \((y_n)\), define \[ \widehat{d}([x_n],[y_n]) := \lim_{n\to\infty} d(x_n,y_n), \] where \([x_n]\) denotes the equivalence class of \((x_n)\).
One must check that the limit exists and does not depend on the chosen representatives. Existence follows because Cauchy sequences ensure that \(d(x_n,y_n)\) is Cauchy in \(\mathbb{R}\): for \(m,n\) large, \[
| d(x_m,y_m)-d(x_n,y_n) | \le d(x_m,x_n)+d(y_m,y_n), |
|---|
\] which becomes small. Independence from representatives uses the equivalence condition \(d(x_n,x_n')\to 0\) and the triangle inequality to show that replacing representatives changes the computed distance by a quantity that vanishes in the limit.
Thus \(\widehat{d}\) is a well-defined metric on \(\widehat{X}\).
2.3 Proving metric space axioms
With \(\widehat{d}\) defined, one verifies:
- Non-negativity: distances are limits of non-negative quantities \(d(x_n,y_n)\).
- Identity of indiscernibles: \(\widehat{d}([x_n],[y_n])=0\) implies \(d(x_n,y_n)\to 0\), so the sequences are equivalent.
- Symmetry: \(\widehat{d}([x_n],[y_n])=\widehat{d}([y_n],[x_n])\) since \(d\) is symmetric.
- Triangle inequality: it comes from applying the triangle inequality to \(d(x_n,z_n)\le d(x_n,y_n)+d(y_n,z_n)\) and taking limits.
These steps ensure \((\widehat{X},\widehat{d})\) is genuinely a metric space.
2.4 Embedding the original space into its completion
2.4.1 Canonical embedding and density
There is a natural map \(\iota:X\to \widehat{X}\): send a point \(x\in X\) to the equivalence class of the constant Cauchy sequence \((x,x,x,\dots)\). This embedding is isometric: \[ \widehat{d}(\iota(x),\iota(y)) = d(x,y). \] Moreover, the image \(\iota(X)\) is dense in \(\widehat{X}\). Given any class \([x_n]\), the metric \(\widehat{d}\big([x_n],\iota(x_k)\big)\) can be made small by choosing \(k\) large, because the tail of \((x_n)\) approximates the would-be limit represented by the class.
Density means that every point in the completion is approximated by points originating from \(X\).
3 Completeness of the resulting space
3.1 Showing every Cauchy sequence converges in the completion
The defining goal is that \(\widehat{X}\) is complete. Take a Cauchy sequence \(\big([x_n^{(m)}]\big)_{m\ge 1}\) in \(\widehat{X}\). Each term is an equivalence class of a Cauchy sequence in \(X\). One constructs a new Cauchy sequence in \(X\) by choosing representatives from each level so that successive terms become close in \(X\). The resulting sequence defines a candidate limit class in \(\widehat{X}\). Then one shows the original Cauchy sequence converges to that class under \(\widehat{d}\).
Conceptually, the completion guarantees that iterated “limit-taking” procedures do not break: Cauchy behavior in \(\widehat{X}\) can be traced back to Cauchy behavior in \(X\), then reassembled into an actual limit point.
3.2 Relating convergence in the completion to Cauchy behavior
For classes \([x_n]\) and \([y_n]\), convergence \([x_n]\to [y_n]\) in \(\widehat{X}\) is equivalent to \[ \widehat{d}([x_n],[y_n]) = \lim_{n\to\infty} d(x_n,y_n) = 0, \] which is exactly the equivalence criterion. More generally, convergence of a sequence in \(\widehat{X}\) aligns with “eventual closeness” between its representing sequences in \(X\).
This correspondence clarifies why the construction is faithful to the metric’s Cauchy structure.
3.3 Uniqueness properties
The completion is unique up to isometry: if \((Y,\rho)\) is another complete metric space receiving an isometric embedding of \(X\) with dense image, then there exists a unique isometry from \(\widehat{X}\) onto \(Y\) that respects the embeddings.
Thus, while the construction depends on choices of representative sequences during proof, the resulting metric space is determined canonically by \(X\) up to the strongest form of metric equivalence.
4 Universal property and functorial aspects
4.1 Uniformly continuous maps into complete spaces
Let \((X,d)\) be a metric space and \((Z,\sigma)\) a complete metric space. If \(f:X\to Z\) is uniformly continuous, then \(f\) carries Cauchy sequences in \(X\) to Cauchy sequences in \(Z\). Completeness of \(Z\) ensures these images converge.
Uniform continuity is the key hypothesis: it controls how \(f\) behaves with respect to small distances, uniformly over all of \(X\), which is required to pass to the limit points created in the completion.
4.2 Extension of maps to the completion
Given \(f:X\to Z\) uniformly continuous, define \(\widehat{f}:\widehat{X}\to Z\) as follows. For a class \([x_n]\), the sequence \(f(x_n)\) converges in \(Z\) (because it is Cauchy), and one sets \[ \widehat{f}([x_n]) := \lim_{n\to\infty} f(x_n). \] One must check that this definition is independent of the representative: if \((x_n)\sim (y_n)\), then \(d(x_n,y_n)\to 0\), and uniform continuity forces \(\sigma(f(x_n),f(y_n))\to 0\), so both limits coincide.
This yields an extension satisfying \(\widehat{f}\circ \iota = f\).
4.3 Cauchy completion as a universal construction
The above extension property characterizes the Cauchy completion in a universal way: it is the “smallest” complete metric space in which uniformly continuous maps out of \(X\) can be extended uniquely.
In practice, this means that many constructions made on \(X\) that are compatible with uniform limits can be performed on \(\widehat{X}\) without re-deriving convergence arguments from scratch.
4.4 Categorical viewpoint (optional)
From a categorical perspective, the completion can be seen as left adjoint to a forgetful-type functor restricted to complete metric spaces with appropriate morphisms. While category-theoretic formulations vary by exact choice of morphisms (uniformly continuous maps, Lipschitz maps, etc.), the core idea is the same: the completion is characterized by an extension/uniqueness condition.
5 Alternative formulations and extensions
5.1 Completion via uniform structures
Every metric space induces a uniform structure, where entourages capture “closeness” in a way compatible with uniform continuity. The Cauchy completion can be reconstructed using Cauchy filters or Cauchy nets in the uniform structure. This approach generalizes the metric construction: it does not require an explicit formula for distance, only the uniform notion of approaching limits.
The metric description is recovered when the uniform structure comes from a metric.
5.2 Completion of pseudometric spaces
If \(d\) is only a pseudometric (allowing \(d(x,y)=0\) for distinct points), the construction proceeds similarly, but the metric on the completion becomes a true metric only after identifying zero-distance points. A common method is: first pass to the metric quotient of \(X\) by the relation \(d(x,y)=0\), then complete the resulting metric space.
This clarifies what changes: the equivalence relation already exists at the level of points in \(X\), not just at the level of Cauchy sequences.
5.3 Completion in normed vector spaces (Banach space context)
| A normed vector space \((V,\|\cdot\|)\) becomes a metric space via \(d(x,y)=\|x-y\|\). Completing in this metric yields a Banach space \(\widehat{V}\): a complete normed vector space into which \(V\) embeds isometrically. |
|---|
Moreover, the vector space operations extend uniquely to \(\widehat{V}\), and the norm agrees with the metric-derived norm. This is central in functional analysis, where one frequently assumes completeness to justify limit arguments.
5.4 Completion in inner product spaces (Hilbert space context)
Similarly, an inner product space has a norm and thus a metric, and completion yields a Hilbert space. The inner product can be extended so that polarization identities remain valid and the resulting inner product reproduces the completed norm.
Hilbert space completeness supports orthogonality, projections, and the convergence properties required for Fourier-type analysis.
6 Examples and computations
6.1 Completion of the rational numbers into the real numbers
| In \(\mathbb{Q}\) with \(d(x,y)= | x-y | \), Cauchy sequences correspond to those sequences that converge in \(\mathbb{R}\). Two Cauchy sequences in \(\mathbb{Q}\) define the same class exactly when they converge to the same real number. |
|---|
Under this identification, the completion \(\widehat{\mathbb{Q}}\) is isometric to \(\mathbb{R}\), with the embedding \(\iota:\mathbb{Q}\to \widehat{\mathbb{Q}}\) corresponding to the usual inclusion of rationals into reals.
6.2 Completion of normed spaces into Banach spaces
Given a normed space \(V\), its completion \(\widehat{V}\) is formed from equivalence classes of Cauchy sequences. The norm on \(\widehat{V}\) arises from \[
| \widehat{\|[x_n]\|} := \lim_{n\to\infty} \|x_n\|. |
|---|
\] Operations like addition and scalar multiplication are defined by representing sequences termwise and passing to equivalence classes. This produces a Banach space where limits of Cauchy sequences exist.
6.3 Completion of discrete and totally bounded spaces
If \(X\) is discrete with metric \(d(x,y)=1\) for \(x\neq y\), then any Cauchy sequence must eventually be constant, hence every discrete metric space is complete. The completion adds nothing new.
For totally bounded spaces, the completion has compactness-related features: the completion of a totally bounded metric space is compact if and only if the original metric space is complete. Thus completion can transform “boundedness of approximations” into genuine compactness by adding missing limits.
7 Relationships to other notions
7.1 Completion versus closure in metric spaces
The closure of a subset \(A\) inside a larger metric space \(B\) is the smallest closed set containing \(A\). Completion, by contrast, constructs a new space where Cauchy sequences acquire limits.
If \(A\subseteq B\), then the completion of \(A\) can often be identified with the closure of \(A\) under suitable assumptions (for example, if the embedding is isometric and \(B\) is complete). Without completeness, closure may still fail to realize all limits of Cauchy sequences that exist in the completion.
7.2 Completion versus taking the metric quotient
Taking the metric quotient collapses points at zero distance, turning a pseudometric into a metric. Completion addresses a different issue: it ensures that Cauchy sequences converge.
One may first quotient to remove degeneracy and then complete; alternatively, one can adapt the completion directly for pseudometric settings, but the resulting structure naturally corresponds to completing after quotienting.
7.3 Links to ultrafilters and Dedekind cuts (briefly)
Other classical limit-completion constructions exist, such as those based on Dedekind cuts for \(\mathbb{Q}\), or structures related to ultrafilters. These approaches encode “limit behavior” in different ways—order cuts or maximal consistent choices—yet they align with the same underlying idea: adding points that represent limits not originally present. For real-number constructions, Dedekind cuts and Cauchy sequences produce equivalent completions.
8 Applications and significance
8.1 Analysis: constructing limits and extending operators
Completion is a foundational tool in analysis. Many theorems assume completeness to ensure that sequences of approximations lead to actual objects. For instance, solving equations or defining operators often begins with an incomplete setting, then completes it so that iterative procedures converge.
The universal property also ensures that uniformly continuous operators extend uniquely to the completed space, preserving structures needed for later arguments.
8.2 Geometry/topology perspectives (as appropriate for metric setting)
In metric geometry, completing a space allows one to interpret “geometric limiting processes” as actual points in a larger space. This helps formalize boundaries, limit behaviors of geodesic-like approximations, and stability of distances under taking limits.
While the completion itself is metric, it interacts naturally with topological notions such as closure and continuity.
8.3 Stability under Cauchy sequences in proofs
Many proof strategies proceed by building a Cauchy sequence of candidates and then using completeness to guarantee convergence. Completion separates these concerns: one can build sequences in the original space, interpret them as points in the completion, and then rely on completeness to conclude existence of limits.
As a result, convergence arguments become more modular: one proves Cauchy-ness, passes to the completion, and then reads off convergence without reworking the underlying space’s deficiencies each time.