Metric Spaces — Open and Closed Sets, Convergence

A metric space $(X, d)$ pairs a set $X$ with a distance function $d$ satisfying positivity, symmetry, and the triangle inequality. Open sets are unions of open balls, closed sets are their complements, and a sequence converges to a limit when its terms eventually stay within every ball around that limit. Used everywhere analysis lives — real numbers, function spaces, fractal geometry. Foundational framework crystallized by Maurice Fréchet (1906) and Felix Hausdorff (1914).

A metric space is the minimal structure needed to make precise the idea of "closeness" and "limit" beyond the real line. The distance function $d : X \times X \to [0, \infty)$ encodes geometry, and from it spring the entire vocabulary of open balls, open and closed sets, interior and closure, and convergence of sequences. This matters because virtually all of analysis — continuity, differentiation, integration, approximation theory — is built atop these notions, and metric spaces let us reuse one set of theorems across $R^{n}$ , spaces of functions, sequence spaces, and more. A surprising fact: the same point set can be open in one metric and not in another, yet many metrics on the same set produce identical collections of open sets — closeness is a topological idea that partly forgets the metric that birthed it.

The Core Insight

Distance generates topology. Once you can measure $d (x, y)$ , you can declare a point "interior" to a set if some ball around it stays inside, declare a set "open" if all its points are interior, and define a limit by "eventually inside every ball." Open and closed sets are not opposites in plain language — they are complement-linked, and a set can be both, neither, or one. Convergence and closedness are two faces of the same coin: closed sets are exactly those that contain all their sequential limits.

Formal Statement

Metric Space

A metric space is a pair $(X, d)$ where $X$ is a set and $d : X \times X \to [0, \infty)$ is a function — the metric — satisfying, for all $x, y, z \in X$ :

d (x, y) = 0 ⟺ x = y

⟳This is the "identity of indiscernibles" axiom; combined with nonnegativity it says distance vanishes exactly on the diagonal.

d (x, y) = d (y, x)

⟳Symmetry — distance does not depend on direction of travel.

d (x, z) \leq d (x, y) + d (y, z)

⟳The triangle inequality; a detour through

y

is never shorter than the direct route.

Here $x, y, z$ are arbitrary points of $X$ . Nonnegativity ( $d \geq 0$ ) follows from the others: $0 = d (x, x) \leq d (x, y) + d (y, x) = 2 d (x, y)$ .

Open Ball

For $x_{0} \in X$ and radius $r > 0$ , the open ball of radius $r$ centered at $x_{0}$ is

B (x_{0}, r) = {x \in X : d (x, x_{0}) < r}

⟳The set of all points strictly closer than

r

; the boundary sphere is excluded by the strict inequality.

The closed ball is $\overset{ˉ}{B} (x_{0}, r) = {x : d (x, x_{0}) \leq r}$ .

Open and Closed Sets

A set $U \subseteq X$ is open if every point is an interior point:

U open ⟺ \forall x \in U, \exists r > 0 such that B (x, r) \subseteq U

⟳Openness = "room to wiggle"; around each member there is a whole ball still inside.

A set $C \subseteq X$ is closed if its complement is open:

C closed ⟺ X ∖ C is open

⟳Closedness is defined by complementation, not directly; this is the cleanest topological definition.

Interior, Closure, Boundary

For $A \subseteq X$ :

int (A) = {x \in A : \exists r > 0, B (x, r) \subseteq A}

⟳The largest open set contained in

A

; collect all interior points.

\overline{A} = {x \in X : \forall r > 0, B (x, r) \cap A \neq = \emptyset}

⟳The closure adds all limit points;

x

is in

\overline{A}

iff every ball around it touches

A

. It equals the smallest closed set containing

A

\partial A = \overline{A} ∖ int (A)

⟳Boundary points have every ball meeting both

A

and its complement.

Convergence of a Sequence

A sequence $(x_{n})_{n \geq 1}$ in $X$ converges to $x \in X$ , written $x_{n} \to x$ , if

\forall ε > 0, \exists N \in N, \forall n \geq N : d (x_{n}, x) < ε

⟳Replace

∣ x_{n} - x ∣

in the real-line definition by

d (x_{n}, x)

; "eventually within every ball."

Equivalently $d (x_{n}, x) \to 0$ as a real sequence.

Topology Axioms from a Metric

Let $(X, d)$ be a metric space and $T$ the collection of open sets. Then:

$\emptyset$ and $X$ are open.
Any union (arbitrary) of open sets is open.
Any finite intersection of open sets is open.

Thus every metric space is a topological space.

Open sets form a topology

Step 1 (trivial sets).

\emptyset

is open vacuously (no points to check).

X

is open since any ball lies in

X

. Step 2 (arbitrary unions). Let

U = ⋃_{i \in I} U_{i}

with each

U_{i}

open. Take

x \in U

; then

x \in U_{i_{0}}

for some

i_{0}

. By openness of

U_{i_{0}}

there is

r > 0

with

B (x, r) \subseteq U_{i_{0}} \subseteq U

. Hence

x

is interior to

U

, so

U

is open. Step 3 (finite intersections). Let

U = ⋂_{k = 1}^{n} U_{k}

, each open. Take

x \in U

. For each

k

there is

r_{k} > 0

with

B (x, r_{k}) \subseteq U_{k}

. Set

r = min {r_{1}, \dots, r_{n}} > 0

(finiteness is essential here). Then

B (x, r) \subseteq B (x, r_{k}) \subseteq U_{k}

for all

k

, so

B (x, r) \subseteq U

. Thus

U

is open.

■

Sequential Characterization of Closed Sets

A set $C \subseteq X$ is closed if and only if it is sequentially closed: whenever $(x_{n}) \subseteq C$ and $x_{n} \to x$ in $X$ , then $x \in C$ .

Closed = contains its limits

( $\Rightarrow$ ) Suppose

C

is closed and

(x_{n}) \subseteq C

with

x_{n} \to x

. If

x \in / C

then

x \in X ∖ C

, which is open, so some

B (x, r) \subseteq X ∖ C

. But

x_{n} \to x

forces

x_{n} \in B (x, r)

for large

n

, contradicting

x_{n} \in C

. Hence

x \in C

. ( $\Leftarrow$ ) Suppose

C

is sequentially closed; we show

X ∖ C

is open. If not, some

x \in X ∖ C

has no ball inside

X ∖ C

: for every

n

B (x, 1/ n) \cap C \neq = \emptyset

. Pick

x_{n} \in B (x, 1/ n) \cap C

. Then

d (x_{n}, x) < 1/ n \to 0

, so

x_{n} \to x

with

(x_{n}) \subseteq C

. By hypothesis

x \in C

, contradicting

x \in X ∖ C

. Hence

X ∖ C

is open and

C

is closed.

■

Open Balls Are Open; Uniqueness of Limits

(a) Every open ball $B (x_{0}, r)$ is an open set. (b) In any metric space, sequential limits are unique: if $x_{n} \to x$ and $x_{n} \to y$ then $x = y$ .

Balls open, limits unique

(a) Let

y \in B (x_{0}, r)

, so

d (y, x_{0}) < r

. Put

s = r - d (y, x_{0}) > 0

. For

z \in B (y, s)

, the triangle inequality [3] gives

d (z, x_{0}) \leq d (z, y) + d (y, x_{0}) < s + d (y, x_{0}) = r

, so

z \in B (x_{0}, r)

. Thus

B (y, s) \subseteq B (x_{0}, r)

and the ball is open. (b) Suppose

x_{n} \to x

and

x_{n} \to y

with

x \neq = y

, so

d (x, y) = δ > 0

. Choose

N

so that

n \geq N

gives

d (x_{n}, x) < δ /2

and

d (x_{n}, y) < δ /2

. Then by [3],

δ = d (x, y) \leq d (x, x_{n}) + d (x_{n}, y) < δ /2 + δ /2 = δ

, a contradiction. So

x = y

■

Maurice Fréchet | 1878–1973

In his 1906 doctoral thesis Sur quelques points du calcul fonctionnel, Fréchet introduced abstract spaces with a distance function — what he called classes (E) — generalizing convergence away from $R^{n}$ to spaces of curves and functions. He recognized that limit and continuity needed only a notion of "écart" (gap), launching general topology and functional analysis. The very word metric space (espace métrique) and the axioms [1]–[3] descend directly from his work. Felix Hausdorff later (1914) refined and popularized these ideas in Grundzüge der Mengenlehre, embedding them in the broader axiomatic study of point-set topology.

Building Intuition

Think of $d$ as a ruler you carry everywhere. An open set is a country with no walls: standing anywhere inside, you can take a small step in any direction and remain in the country. A closed set may include its border fence — points on the edge belong to it. The number line illustrates the spectrum: $(0, 1)$ is open (every point has wiggle room), $[0, 1]$ is closed (it owns its endpoints), $[0, 1)$ is neither, and $\emptyset$ and $R$ are simultaneously both open and closed.

Convergence is a "settling down" idea: no matter how tiny a target ball you draw around the limit, the sequence eventually crawls inside and never leaves. Crucially, which sets are open depends on the metric, but two different metrics can declare the same open sets (then they are topologically equivalent), in which case they agree on every convergence, continuity, and closedness question.

Open vs. closed: a point interior to an open set has a full ball inside, while a boundary point of a closed set sits on the rim

Worked Examples

Convergence in the Euclidean plane

Given:

X = R^{2}

with Euclidean metric

d (p, q) = (p_{1} - q_{1})^{2} + (p_{2} - q_{2})^{2}

. Consider

x_{n} = (\frac{1}{n}, 2 - \frac{1}{n ^{2}})

. Does it converge, and to what? Step 1 — guess the limit. As

n \to \infty

1/ n \to 0

and

2 - 1/ n^{2} \to 2

. Candidate limit

x = (0, 2)

. Step 2 — compute the distance.

d (x_{n}, x) = (\frac{1}{n})^{2} + (\frac{1}{n ^{2}})^{2} = \frac{1}{n ^{2}} + \frac{1}{n ^{4}} .

Step 3 — bound it. Since

1/ n^{4} \leq 1/ n^{2}

, we get

d (x_{n}, x) \leq 2/ n^{2} = 2 / n

. Step 4 — apply the definition [10]. Given

ε > 0

, choose

N > 2 / ε

. Then

n \geq N \Rightarrow d (x_{n}, x) \leq 2 / n \leq 2 / N < ε

. Answer:

x_{n} \to (0, 2) in (R^{2}, d_{Euclidean})

Closure and interior of a rational interval

Given:

X = R

with

d (x, y) = ∣ x - y ∣

. Let

A = (1, 3] \cap Q

(rationals in the interval). Step 1 — interior. Any open ball

B (x, r) = (x - r, x + r)

contains irrationals, which are not in

A

. So no point of

A

has a ball inside

A

int (A) = \emptyset .

Step 2 — closure. Every ball around any

x \in [1, 3]

meets the rationals of

(1, 3]

(rationals are dense). Points outside

[1, 3]

have balls disjoint from

A

\overline{A} = [1, 3] .

Step 3 — boundary. By [9],

\partial A = \overline{A} ∖ int (A) = [1, 3] ∖ \emptyset = [1, 3]

. Answer:

int (A) = \emptyset, \overline{A} = [1, 3], \partial A = [1, 3]

A set can have empty interior yet a "fat" closure — density makes the difference.

Common Misconceptions

"Closed means not open." False. Open and closed are not complementary properties. In any metric space

\emptyset

and

X

are both open and closed (clopen). The half-open interval

[0, 1)

is neither. The correct relationship is via complements [6]:

C

is closed iff

X ∖ C

is open. "Infinite intersections of open sets are open." False. In

R

⋂_{n \geq 1} (- 1/ n, 1/ n) = {0}

, a closed set, not open. Only finite intersections are guaranteed open (see [Step 3] of the topology proof). "The closed ball equals the closure of the open ball." Not always! In a general metric space

\overline{B (x, r)} \subseteq \overset{ˉ}{B} (x, r)

, but equality can fail.

Closed ball strictly larger than closure of open ball

Take $X$ to be any set with two or more points equipped with the discrete metric

d (x, y) = {0, 1, x = y x \neq = y .

Fix a point

p

and radius

r = 1

Open ball: $B (p, 1) = {x : d (x, p) < 1} = {p}$ . Its closure is $\overline{{p}} = {p}$ (singletons are closed here).
Closed ball: $\overset{ˉ}{B} (p, 1) = {x : d (x, p) \leq 1} = X$ (everything is within distance $1$ ).

So $\overline{B (p, 1)} = {p} ⊊ X = \overset{ˉ}{B} (p, 1)$ . The naive intuition from $R^{n}$ breaks. (In the discrete metric every subset is both open and closed, and a sequence converges iff it is eventually constant.)

Connections

The open/closed/convergence machinery is the gateway to the rest of analysis. Continuity is characterized cleanly here: $f$ is continuous iff preimages of open sets are open, iff $x_{n} \to x \Rightarrow f (x_{n}) \to f (x)$ — see Continuous Functions on Metric Spaces. Completeness asks whether Cauchy sequences converge, refining the convergence notion of [10] into Complete Metric Spaces and powering the Banach Fixed-Point Theorem. Compactness (Compact Metric Spaces) combines closedness with a boundedness-like finiteness via sequential or open-cover formulations. The whole framework abstracts to Topological Spaces, where open sets are taken as primitive and the metric is forgotten. Finally, different metrics yielding the same open sets connect to Equivalent Metrics and Norms.

Real-World Applications

Signal & image processing — function spaces like $L^{2}$ use the metric $d (f, g) = ∥ f - g ∥_{2}$ ; convergence of Fourier partial sums and JPEG compression error are statements about closeness in this metric.
Machine learning / k-nearest neighbors — classification literally computes distances; choosing Euclidean vs. Manhattan ( $ℓ^{1}$ ) vs. cosine metrics changes which "balls" group data points.
Numerical analysis — iterative solvers (Newton's method, Gauss–Seidel) are proven convergent by showing iterates form a convergent (often Cauchy) sequence in a suitable metric.
GPS and routing — road networks form metric spaces (shortest-path metric); open balls are "everything reachable within distance $r$ ," used in isochrone maps.
Genomics — the edit distance (Levenshtein) on DNA strings is a genuine metric; sequence alignment and clustering rely on its triangle inequality.
Fractal compression & dynamics — the Hausdorff metric on compact sets makes the space of fractals a complete metric space, where iterated function systems converge to attractors.
Control theory — stability of a system means trajectories converge to an equilibrium in state-space metric, formalizing Lyapunov analysis.

Historical Development

1817 — Bolzano gives an early rigorous notion of continuity and limit, foreshadowing the $ε$ -language.
1872 — Weierstrass and Heine codify the $ε$ – $δ$ / $ε$ – $N$ definitions of continuity and convergence on $R$ .
1872 — Cantor and Dedekind construct the reals; Cantor introduces accumulation (limit) points and derived sets, the seed of "closure."
1906 — Maurice Fréchet's thesis introduces abstract metric spaces (classes (E)) and the axioms of distance, generalizing convergence to function spaces.
1910 — Fréchet and F. Riesz develop early functional-analytic spaces; the notion of completeness in abstract settings emerges.
1914 — Felix Hausdorff's Grundzüge der Mengenlehre axiomatizes neighborhoods and topological spaces, placing metric spaces inside a wider theory; the Hausdorff separation axiom is born.
1922 — Stefan Banach's thesis founds normed/complete spaces (Banach spaces) and proves the contraction mapping theorem.
1906–1930 — The Moscow and Polish (Lwów–Warsaw) schools (Urysohn, Alexandroff, Kuratowski, Sierpiński) systematize open/closed sets, closure operators (Kuratowski axioms), and metrization theorems.
1933 — Urysohn's and later metrization theorems characterize exactly which topological spaces arise from a metric, closing the abstraction loop.

Subtleties & Limits

While metric spaces are well-understood, their boundaries spark ongoing research. Metrizability: precisely which topological spaces admit a compatible metric is governed by the Nagata–Smirnov and Bing metrization theorems, but generalizations to non-Hausdorff or quasi-metric settings remain subtle. Coarse geometry & large-scale structure: Gromov's program studies metric spaces "up to bounded distortion," central to geometric group theory and still generating open questions about embeddings of groups into Hilbert space (the Novikov and coarse Baum–Connes conjectures). Metric embedding theory (active in theoretical computer science) asks how faithfully arbitrary finite metrics embed into $ℓ^{p}$ spaces with low distortion — Bourgain's theorem gives $O (lo g n)$ , and optimal distortions for specific families stay open. Wasserstein / optimal-transport metrics on probability measures are an explosively active area with deep open questions in geometry and PDE. Even the innocent fact that closed balls can differ from closures of open balls signals how much geometric information a bare metric retains — and how much it discards.

Metric Spaces — Open and Closed Sets, Convergence

The Core Insight

Formal Statement

Metric Space

Open Ball

Open and Closed Sets

Interior, Closure, Boundary

Convergence of a Sequence

Topology Axioms from a Metric

Open sets form a topology

Sequential Characterization of Closed Sets

Closed = contains its limits

Open Balls Are Open; Uniqueness of Limits

Balls open, limits unique

Maurice Fréchet | 1878–1973

Building Intuition

Open vs. closed: a point interior to an open set has a full ball inside, while a boundary point of a closed set sits on the rim

Worked Examples

Convergence in the Euclidean plane

Closure and interior of a rational interval

Common Misconceptions

Closed ball strictly larger than closure of open ball

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also

Before this

Where this leads