1 Definition and motivation
1.1 Cauchy sequences in metric spaces
A metric space \((X,d)\) provides a notion of distance between points. A sequence \((x_n)\subseteq X\) is called a Cauchy sequence if, for every \(\varepsilon>0\), there exists \(N\) such that \(d(x_m,x_n)<\varepsilon\) whenever \(m,n\ge N\). Informally, the terms of the sequence become arbitrarily close to each other as the index grows, regardless of whether the sequence converges to a point in \(X\).
Cauchy sequences are central because they capture “eventual stability” in a space, even when the space may be incomplete and fail to contain the limiting points.
1.2 Equivalence relation based on vanishing differences
Given a metric space \((X,d)\), one common definition introduces an equivalence relation on the set of Cauchy sequences. Two Cauchy sequences \((x_n)\) and \((y_n)\) are declared equivalent if \[ d(x_n,y_n)\to 0 \quad \text{as } n\to\infty. \] This criterion reflects that the two sequences eventually approach each other at vanishing distance, so they represent the same limiting behavior.
1.3 Equivalence classes as “same limit behavior”
The equivalence class of a Cauchy sequence \((x_n)\) consists of all Cauchy sequences that become indistinguishable at the level of distance in the limit. Even if \((x_n)\) does not converge inside \(X\), its equivalence class can still be treated as a new object representing the would-be limit.
This viewpoint turns “approach without convergence” into a precise algebraic set-theoretic construction.
1.4 Examples of equivalence classes
- Convergent sequences in a complete space. If \(X\) is complete and \((x_n)\to x\), then \((x_n)\) is equivalent to the constant sequence \((x,x,x,\dots)\), because \(d(x_n,x)\to 0\).
- Different sequences with the same limit behavior in an incomplete space. In an incomplete space, two sequences may fail to converge there but still satisfy \(d(x_n,y_n)\to 0\). Their equivalence class represents the same “missing limit.”
- Rational approximations. Sequences of rationals that approximate an irrational number, such as decimal truncations of \(\sqrt{2}\), form Cauchy sequences. Distinct approximation methods yield sequences equivalent under the vanishing difference condition.
2 Properties of the equivalence relation
2.1 Reflexivity
For any Cauchy sequence \((x_n)\), the distance between corresponding terms satisfies \(d(x_n,x_n)=0\) for all \(n\). Hence \(d(x_n,x_n)\to 0\), so the sequence is equivalent to itself.
2.2 Symmetry
If \((x_n)\sim (y_n)\), then \(d(x_n,y_n)\to 0\). By symmetry of the metric, \(d(y_n,x_n)=d(x_n,y_n)\to 0\), so \((y_n)\sim (x_n)\).
2.3 Transitivity
Suppose \((x_n)\sim (y_n)\) and \((y_n)\sim (z_n)\). Then \(d(x_n,y_n)\to 0\) and \(d(y_n,z_n)\to 0\). The triangle inequality gives \[ d(x_n,z_n)\le d(x_n,y_n)+d(y_n,z_n)\to 0, \] so \((x_n)\sim (z_n)\).
2.4 Dependence on the underlying metric
The equivalence relation is defined using the metric \(d\). Changing the metric changes which sequences are considered Cauchy and changes the meaning of “distance between corresponding terms.” As a result, equivalence classes are not intrinsic to the set \(X\) alone; they depend on the metric structure.
3 Constructing elements from Cauchy sequences
3.1 Quotient set of Cauchy sequences
Let \(\mathcal{C}(X)\) denote the set of all Cauchy sequences in \((X,d)\). The set of equivalence classes is the quotient \[ \widehat{X} := \mathcal{C}(X)/\sim, \] where \(\sim\) is the equivalence relation based on \(d(x_n,y_n)\to 0\). Each element of \(\widehat{X}\) corresponds to one “limit behavior” among Cauchy sequences.
3.2 Canonical representative viewpoint
Although elements of \(\widehat{X}\) are equivalence classes, computations often use representatives. A typical abuse of language is to write an element of \(\widehat{X}\) as if it were a sequence \((x_n)\), with the understanding that different equivalent sequences produce the same element. This representative-based perspective must always respect the equivalence relation.
3.3 Embedding original points as constant sequences
There is a natural map \(i:X\to \widehat{X}\) defined by sending a point \(x\in X\) to the equivalence class of the constant sequence \((x,x,x,\dots)\). Since constant sequences are Cauchy, this map is well-defined.
When \(X\) is a metric space, this embedding preserves the distance structure in a precise way after the metric on \(\widehat{X}\) is defined.
3.4 Well-definedness of induced operations
To define operations on \(\widehat{X}\), one typically specifies how to combine representatives (e.g., addition term-by-term) and then checks that the resulting equivalence class does not depend on the choice of representatives. If \((x_n)\sim (x_n')\) and \((y_n)\sim (y_n')\), then one must show that the constructed sequences satisfy the same equivalence. This “compatibility with the quotient” is the key technical requirement for operations to be meaningful on equivalence classes.
4 Metric and completeness on equivalence classes
4.1 Defining a distance between classes
A standard metric on \(\widehat{X}\) is defined as follows. For equivalence classes \([x_n]\) and \([y_n]\), \[ \widehat{d}([x_n],[y_n]) := \lim_{n\to\infty} d(x_n,y_n), \] provided the limit exists.
For Cauchy sequences, the limit exists because the sequence \(d(x_n,y_n)\) is Cauchy in \(\mathbb{R}\). One can verify this using the triangle inequality and the fact that both \((x_n)\) and \((y_n)\) are Cauchy.
4.1.1 Independence from chosen representatives
If \((x_n)\sim (x_n')\) and \((y_n)\sim (y_n')\), then \(d(x_n,x_n')\to 0\) and \(d(y_n,y_n')\to 0\). The triangle inequality implies \[
| \big | d(x_n,y_n)-d(x_n',y_n')\big |
|---|
\le d(x_n,x_n') + d(y_n,y_n') \to 0. \] Thus \(\lim d(x_n,y_n)=\lim d(x_n',y_n')\), so \(\widehat{d}\) does not depend on the representatives.
4.2 Proving the metric axioms
Using \(\widehat{d}\), one checks:
- Nonnegativity: distances are always \(\ge 0\) since \(d\ge 0\).
- Identity of indiscernibles: \(\widehat{d}([x_n],[y_n])=0\) implies \(\lim d(x_n,y_n)=0\), so \((x_n)\sim (y_n)\), meaning the classes are equal.
- Symmetry: \(\widehat{d}([x_n],[y_n])=\widehat{d}([y_n],[x_n])\) because \(d(x_n,y_n)=d(y_n,x_n)\).
- Triangle inequality: follows from applying the triangle inequality for \(d\) term-by-term and then passing to limits.
4.3 Showing completeness
The construction is designed so that \(\widehat{X}\) becomes complete. Concretely, if \(([x_n^{(k)}])_{k\ge 1}\) is a Cauchy sequence in \(\widehat{X}\), one can diagonalize to build a representative sequence in \(X\) whose equivalence class serves as the limit in \(\widehat{X}\). The metric definition ensures that this limit is consistent with the Cauchy behavior of the original sequence of classes.
4.4 Identifying limits via Cauchy representatives
Limits in \(\widehat{X}\) can be described in terms of Cauchy representatives: if \([x_n]\) represents an element, then convergence in \(\widehat{X}\) corresponds to the distance between representatives tending to zero: \[ [x_n^{(k)}] \to [x_n] \quad \Longleftrightarrow \quad \widehat{d}([x_n^{(k)}],[x_n])\to 0. \] This allows one to translate convergence questions about new elements back to distance statements about sequences from the original space.
5 Algebraic structure (when applicable)
5.1 Vector space structures induced by sequences
If \(X\) carries an algebraic structure, such as being a vector space over \(\mathbb{R}\) or \(\mathbb{C}\), and the metric is compatible with that structure (e.g., comes from a norm), then operations can be induced on equivalence classes. Typically, addition is defined by \[ [x_n] + [y_n] := [x_n+y_n], \] and scalar multiplication by \[ \alpha [x_n] := [\alpha x_n]. \] Well-definedness requires that if \((x_n)\sim (x_n')\) and \((y_n)\sim (y_n')\), then the sequences \((x_n+y_n)\) and \((x_n'+y_n')\) are equivalent.
5.2 Compatibility with scalar multiplication
| When the metric is induced by a norm, such as \(d(u,v)=\|u-v\|\), the key property used for well-definedness is continuity of algebraic operations with respect to the norm. If \(\|x_n-x_n'\|\to 0\), then \(\|\alpha x_n-\alpha x_n'\|= | \alpha | \,\|x_n-x_n'\|\to 0\). Similar reasoning applies to addition. |
|---|
5.3 Multiplication and ring/field structure (special cases)
If \(X\) is a normed algebra (e.g., a normed ring or a normed field) and the multiplication is continuous in a suitable sense, one may define \[ [x_n]\cdot [y_n] := [x_n y_n]. \] Well-definedness can require additional bounds relating products to norms. Under common conditions (such as submultiplicativity of the norm), the completion inherits a corresponding algebraic structure, turning it into a completed normed algebra or completed field.
6 Universal property and completion perspective
6.1 Completion as a minimal complete extension
The completion \(\widehat{X}\) is not merely a set; it is characterized by a universal property. Roughly, it is a complete metric space that contains \(X\) densely and admits extensions of uniformly continuous maps from \(X\) to any other complete metric space.
In this sense, completion is the “smallest” complete object that realizes the Cauchy behavior present in \(X\).
6.2 Isometries that respect equivalence classes
Any isometry (or more generally a uniformly continuous map) from \(X\) into a complete metric space \(Y\) can be extended to a map from \(\widehat{X}\) to \(Y\). The extension is determined by how images of Cauchy sequences behave: if \(f:X\to Y\) preserves distances, then applying \(f\) term-by-term to a Cauchy sequence yields a Cauchy sequence in \(Y\), which converges because \(Y\) is complete.
6.3 Uniqueness up to isomorphism
The completion is unique up to canonical isometric isomorphism. If two complete metric spaces both complete \(X\) in a way that respects the embedding, there exists an isometry between them aligning corresponding elements. This uniqueness means the construction does not depend on arbitrary choices.
7 Standard examples and computations
7.1 Construction of the real numbers from rationals
| The real numbers \(\mathbb{R}\) can be obtained by completing the metric space \((\mathbb{Q}, | \cdot | )\). Equivalence classes consist of Cauchy sequences of rationals modulo the relation \( | x_n-y_n | \to 0\). Under this identification, every rational number corresponds to the class of the constant sequence, and every irrational real number corresponds to the class of a Cauchy rational approximation that does not converge in \(\mathbb{Q}\). |
|---|
7.2 Irrational numbers via Cauchy sequences
An irrational number such as \(\sqrt{2}\) can be represented by a Cauchy sequence of rationals, for example the sequence of decimal truncations or rational approximations produced by a numerical method. Different approximation procedures yield sequences whose pairwise differences tend to zero, and therefore they determine the same equivalence class.
This correspondence is what makes the completed space genuinely larger than the original one while still reflecting “limit behavior” accurately.
7.3 Comparing representations of the same real value
To show two Cauchy sequences represent the same real value, one checks the equivalence condition. Given Cauchy sequences \((a_n)\) and \((b_n)\) in \(\mathbb{Q}\), one verifies whether \[
| a_n-b_n | \to 0. |
|---|
\] If this holds, then \([a_n]=[b_n]\) in the completion. This criterion is often used in proofs where two different constructions of a real number are shown to agree.
8 Common pitfalls and clarifications
8.1 Confusing convergence with being Cauchy
A sequence can be Cauchy without converging in an incomplete space. Completeness is precisely the property that every Cauchy sequence converges. Confusing these notions can lead to incorrect assumptions about whether limits exist within the original space.
8.2 Incorrect equivalence notions
The equivalence relation must match the intended “distance vanishing” criterion. For instance, declaring equivalence by eventual equality (which is too strict) or by convergence of termwise differences in a different topology (which may not align with the metric) can produce objects that do not correspond to the desired completion.
8.3 Representative dependence and how to avoid it
Many definitions in the completion use representatives. The central safeguard is to prove well-definedness: any quantity defined on representatives must be shown to yield the same result for equivalent sequences. If that check is omitted, a supposed element or operation may depend on arbitrary choices rather than on the equivalence class itself.
9 Related concepts
9.1 Quotient constructions in analysis
The formation of equivalence classes resembles other quotient constructions in analysis, such as identifying functions that differ on sets of measure zero. In both cases, the goal is to treat objects that are “indistinguishable” under a chosen criterion as the same.
9.2 Completion of normed spaces
| Replacing a metric by a norm \( \|\cdot\| \) produces the standard setting of normed spaces. The completion of a normed space yields a Banach space, where the metric comes from the norm. The same equivalence-class construction underlies this process. |
|---|
9.3 Connections to sequences, limits, and metrics
The completion framework ties together three foundational ideas: Cauchy sequences capture internal consistency; equivalence classes identify sequences with the same asymptotic behavior; and the induced metric makes these abstract “limits” into a geometric object where convergence becomes available through completeness.