Metric Spaces: Distance, Convergence, Completeness

⚡ Key Takeaways

A metric space is a set equipped with a distance function satisfying four axioms. Convergence means points get arbitrarily close to a limit; completeness means every "self-approaching" (Cauchy) sequence actually has a limit inside the space. Core result: $\mathbb{R}^n$ is complete, but $\mathbb{Q}$ is not. These ideas underpin analysis, topology, numerical methods, and functional analysis. Foundational concepts formalized by Maurice Fréchet (1906) and Felix Hausdorff (1914).

💡 The Core Insight

Distance is not a number attached to space — it is a function with axioms. Once you have a valid distance function, the entire machinery of limits, open sets, and continuity follows automatically and uniformly. Completeness is the property that distinguishes spaces where "convergence behaves" (no holes) from those riddled with gaps. The Cauchy criterion lets you certify convergence without knowing the limit in advance — but only in complete spaces does Cauchy-ness guarantee a limit exists.

DefinitionMetric Space

A metric space is a pair $(X,d)$ where $X$ is a nonempty set and $d:X\times X\to\mathbb{R}$ is a function (the metric or distance) satisfying, for all $x,y,z\in X$ :

1

\quad d(x,y)\ge 0,\qquad d(x,y)=0 \iff x=y

Derivation hint: the first part is non-negativity; the second (identity of indiscernibles) forces distinct points to have positive distance.

2

\quad d(x,y)=d(y,x)

Derivation hint: symmetry — distance does not depend on direction of measurement.

3

\quad d(x,z)\le d(x,y)+d(y,z)

Derivation hint: the triangle inequality — a detour through $y$ is never shorter than the direct route. Here $X$ is the underlying set, $x,y,z$ are points, and $d(x,y)$ is the distance between $x$ and $y$ .

DefinitionOpen Ball, Open and Closed Sets

For $x_0\in X$ and radius $r>0$ , the open ball is

4

\quad B(x_0,r)=\{x\in X : d(x,x_0)<r\}

Derivation hint: collect all points strictly within distance $r$ of the center. A set $U\subseteq X$ is open if every point of $U$ has some ball entirely inside $U$ ; a set is closed if its complement is open. These generate the metric topology.

DefinitionConvergence of a Sequence

A sequence $(x_n)_{n\ge1}$ in $(X,d)$ converges to $L\in X$ , written $x_n\to L$ , if

5

\quad \forall\varepsilon>0\ \ \exists N\in\mathbb{N}\ \ \forall n\ge N:\ d(x_n,L)<\varepsilon

Derivation hint: distance to the limit can be forced below any threshold $\varepsilon$ from some index $N$ onward. The limit, when it exists, is unique (a consequence of the triangle inequality).

DefinitionCauchy Sequence

A sequence $(x_n)$ is Cauchy if its terms eventually cluster together:

6

\quad \forall\varepsilon>0\ \ \exists N\in\mathbb{N}\ \ \forall m,n\ge N:\ d(x_m,x_n)<\varepsilon

Derivation hint: instead of comparing to a limit (which may not be known), compare terms to each other.

DefinitionComplete Metric Space

$(X,d)$ is complete if every Cauchy sequence in $X$ converges to a point in $X$ .

TheoremConvergent ⟹ Cauchy

Every convergent sequence in a metric space is Cauchy. The converse holds iff the space is complete.

TheoremUniqueness of Limits

In any metric space, if $x_n\to L$ and $x_n\to L'$ , then $L=L'$ .

ProofConvergent sequences are Cauchy

Step 1. Suppose $x_n\to L$ . Fix $\varepsilon>0$ .

Step 2. By [5] applied with $\varepsilon/2$ , there exists $N$ such that for all $n\ge N$ , $d(x_n,L)<\varepsilon/2$ .

Step 3. Take any $m,n\ge N$ . By the triangle inequality [3],

7

\quad d(x_m,x_n)\le d(x_m,L)+d(L,x_n)<\tfrac{\varepsilon}{2}+\tfrac{\varepsilon}{2}=\varepsilon

Derivation hint: route the comparison through the common limit $L$ ; symmetry [2] lets you write $d(L,x_n)=d(x_n,L)$ .

Step 4. Since $\varepsilon$ was arbitrary, [6] holds, so $(x_n)$ is Cauchy. $\blacksquare$

∎

ProofUniqueness of limits

Step 1. Suppose $x_n\to L$ and $x_n\to L'$ . By the triangle inequality,

8

\quad 0\le d(L,L')\le d(L,x_n)+d(x_n,L')

Derivation hint: insert $x_n$ between $L$ and $L'$ via [3].

Step 2. As $n\to\infty$ , both terms on the right tend to $0$ by [5]. Hence $d(L,L')\le 0$ .

Step 3. By non-negativity [1], $d(L,L')=0$ , so $L=L'$ . $\blacksquare$

∎

mathematician · Maurice Fréchet | 1878–1973

In his 1906 doctoral thesis Sur quelques points du calcul fonctionnel, Fréchet introduced the abstract notion of an "écart" (gap) and the concept of espaces métriques — the first systematic treatment of distance as an axiomatic structure independent of $\mathbb{R}^n$ . He also defined compactness via sequences and laid groundwork for functional analysis. Felix Hausdorff later popularized and refined these ideas in his 1914 Grundzüge der Mengenlehre, coining the now-standard "metric space" terminology.

🧠 Intuition

Think of a metric as a rulebook for nearness rather than a literal ruler. The triangle inequality is the heart: it forbids "shortcuts," ensuring distances behave coherently. Convergence is "homing in" on a target. Cauchy is "the swarm of terms shrinking toward a single point" — but the swarm might be shrinking toward a hole not contained in your space. Completeness is the guarantee there are no holes: every place the terms are heading toward is actually there. The classic mental image: $\mathbb{Q}$ is a number line riddled with invisible punctures (the irrationals), and a Cauchy sequence of rationals can march straight into a puncture and vanish. $\mathbb{R}$ fills in every puncture.

Figure

A Cauchy sequence converging in ℝ but escaping through a "hole" in ℚ

ExampleVerifying a metric and a limit in ℝ²

Given: $X=\mathbb{R}^2$ with the Euclidean metric $d(\mathbf{x},\mathbf{y})=\sqrt{(x_1-y_1)^2+(x_2-y_2)^2}$ . Consider the sequence $\mathbf{x}_n=\left(\dfrac{1}{n},\, 2-\dfrac{1}{n^2}\right)$ .

Step 1 — Candidate limit. Componentwise, $1/n\to 0$ and $2-1/n^2\to 2$ , so guess $L=(0,2)$ .

Step 2 — Compute distance.

9

\quad d(\mathbf{x}_n,L)=\sqrt{\left(\tfrac1n\right)^2+\left(\tfrac{1}{n^2}\right)^2}=\sqrt{\tfrac1{n^2}+\tfrac1{n^4}}

Derivation hint: subtract components, square, and apply the Euclidean formula.

Step 3 — Bound it. For $n\ge1$ , $\dfrac{1}{n^4}\le\dfrac{1}{n^2}$ , so $d(\mathbf{x}_n,L)\le\sqrt{\tfrac{2}{n^2}}=\dfrac{\sqrt2}{n}$ .

Step 4 — Apply definition. Given $\varepsilon>0$ , choose $N>\sqrt2/\varepsilon$ . Then for $n\ge N$ , $d(\mathbf{x}_n,L)\le\sqrt2/n<\varepsilon$ .

\boxed{\mathbf{x}_n\to(0,\,2)\ \text{in}\ (\mathbb{R}^2,d_{\text{Euclidean}})}

ExampleA Cauchy sequence that fails to converge in ℚ

Given: $X=\mathbb{Q}$ with $d(x,y)=|x-y|$ . Define $x_1=1$ and $x_{n+1}=\dfrac{x_n}{2}+\dfrac{1}{x_n}$ (Newton's iteration for $\sqrt2$ ). All $x_n\in\mathbb{Q}$ .

Step 1 — It is Cauchy. The iteration converges quadratically to $\sqrt2$ ; numerically $x_2=1.5$ , $x_3\approx1.41667$ , $x_4\approx1.414216$ . The differences shrink geometrically:

10

\quad |x_{n+1}-x_n|\to 0\ \text{rapidly},\quad\text{so}\quad \forall\varepsilon>0\ \exists N:\ |x_m-x_n|<\varepsilon\ (m,n\ge N)

Derivation hint: terms agree to more decimal places each step, satisfying [6].

Step 2 — Its only possible limit. If $x_n\to L$ in $\mathbb{R}$ , taking limits gives $L=\tfrac{L}{2}+\tfrac1L\Rightarrow L^2=2\Rightarrow L=\sqrt2$ .

Step 3 — Conclusion in ℚ. Since $\sqrt2\notin\mathbb{Q}$ , there is no $L\in\mathbb{Q}$ with $x_n\to L$ . The sequence is Cauchy but does not converge in $\mathbb{Q}$ .

\boxed{(\mathbb{Q},|\cdot|)\ \text{is NOT complete; the same sequence converges to}\ \sqrt2\ \text{in}\ \mathbb{R}}

⚠️ Common mistake

"Cauchy means convergent." False in general — only in complete spaces. The implication convergent ⟹ Cauchy always holds; the reverse requires completeness. Always state which space you are working in.

"Every distance comes from a norm." No. The discrete metric $d(x,y)=1$ for $x\ne y$ is a perfectly valid metric on any set, even one with no algebraic structure, and it comes from no norm.

"Closed sets are the same as complete sets." Different notions. A closed subset of a complete space is complete; but "closed" is relative to an ambient space, while "complete" is intrinsic.

"The triangle inequality is optional decoration." It is the load-bearing axiom: uniqueness of limits, continuity of $d$ , and the well-definedness of the topology all depend on it.

CounterexampleCompleteness depends on the metric, not just the set

Take $X=(0,1)$ with the standard metric $|\cdot|$ . The sequence $x_n=1/n$ is Cauchy but converges to $0\notin(0,1)$ — so $(0,1)$ is not complete. Yet the same set $(0,1)$ can be made complete by a different metric. Define $\rho(x,y)=\left|\tan\!\big(\pi(x-\tfrac12)\big)-\tan\!\big(\pi(y-\tfrac12)\big)\right|$ , which transports the complete metric of $\mathbb{R}$ onto $(0,1)$ . Under $\rho$ , the sequence $1/n$ is not even Cauchy. Moral: completeness is a property of the pair $(X,d)$ , never of the bare set $X$ .

connection

Metric spaces sit between general Topological Spaces (where only open sets, not distances, are given) and Normed Vector Spaces (where distance comes from a norm via $d(x,y)=\|x-y\|$ ). A complete normed space is a Banach Space; a complete inner-product space is a Hilbert Space. The notion of Continuity becomes the $\varepsilon$ – $\delta$ condition transcribed into $d$ . Compactness in metric spaces has the elegant sequential characterization (every sequence has a convergent subsequence), and a metric space is compact iff it is complete and totally bounded. The Banach Fixed-Point Theorem lives entirely on complete metric spaces. Finally, the Completion of a Metric Space (filling in the holes via equivalence classes of Cauchy sequences) is the abstract engine behind the Cantor construction of The Real Numbers.

🔬 Where this appears

⏳ History

- 1817 — Bolzano gives an early statement of the Cauchy criterion for real sequences.

- 1821 — Cauchy, in Cours d'analyse, formalizes the convergence criterion now bearing his name.

- 1872 — Cantor and Dedekind (independently) construct

\mathbb{R}

, Cantor explicitly via Cauchy sequences of rationals — the prototype of completion.

- 1906 — Maurice Fréchet's thesis introduces abstract espaces métriques and the axioms of distance, founding metric space theory.

- 1914 — Felix Hausdorff's Grundzüge der Mengenlehre systematizes metric and topological spaces and standardizes the terminology.

- 1922 — Stefan Banach proves the contraction mapping (fixed-point) theorem on complete metric spaces in his doctoral thesis, and develops complete normed spaces (Banach spaces).

- 1906–1930s — Riesz, Fischer, and others establish completeness of

L^p

spaces (Riesz–Fischer theorem), central to functional analysis.

- 1937 — Kuratowski and others develop the abstract theory of completion and metrization conditions.

- Mid–20th c. — Metric methods diffuse into probability (Prokhorov, Wasserstein metrics), geometry (Gromov–Hausdorff distance), and computer science.

open-problem

Several subtle frontiers persist. Metrizability: not every topological space arises from a metric; the Urysohn and Nagata–Smirnov metrization theorems characterize exactly which do, but checking these conditions can be delicate. Choice of metric matters profoundly: homeomorphic spaces may differ in completeness, and finding a complete metric inducing a given topology (when one exists — such spaces are called completely metrizable or Polish if also separable) is a nontrivial classification problem central to descriptive set theory. Infinite-dimensional pathologies: in infinite-dimensional Banach spaces, closed bounded sets need not be compact (failure of Heine–Borel), so completeness alone does not recover finite-dimensional intuition. Generalizations: weakening the axioms yields pseudometrics (drop $d(x,y)=0\Rightarrow x=y$ ), quasimetrics (drop symmetry, relevant in computer science and asymmetric travel costs), and ultrametrics ( $d(x,z)\le\max(d(x,y),d(y,z))$ , governing $p$ -adic numbers and phylogenetics) — each with its own active theory.

Metric Spaces: Distance, Convergence, Completeness

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also