Compactness and the Heine-Borel Theorem

Compactness is a topological property that generalizes "closed and bounded." The Heine–Borel Theorem states that a subset of

R^{n}

is compact if and only if it is closed and bounded. Applies throughout real analysis, topology, and optimization wherever existence of maxima, uniform continuity, or convergent subsequences is needed. Named after Eduard Heine (who used the underlying idea c. 1872) and Émile Borel (who proved the covering form in 1895).

A subset $K$ of a topological space is compact if every open cover of $K$ admits a finite subcover. In metric spaces this is equivalent to sequential compactness (every sequence has a convergent subsequence with limit in $K$ ) and to being complete and totally bounded. Compactness matters because it converts infinite, potentially unmanageable covering data into finite data, allowing one to extract uniform conclusions — continuous functions attain their extrema, are uniformly continuous, and so on. The surprising fact is that in $R^{n}$ this abstract finiteness property collapses to the elementary condition "closed and bounded," yet in infinite-dimensional spaces this collapse fails dramatically: the closed unit ball is bounded and closed but not compact.

The Core Insight

Compactness is finiteness in disguise. It says you can always replace "infinitely many open sets covering $K$ " with "finitely many of them." This single trick lets local information (defined patchwise on the cover) be glued into global, uniform information. Heine–Borel identifies exactly which subsets of Euclidean space enjoy this trick: precisely the closed and bounded ones.

Formal Statement

Open Cover and Subcover

Let $K$ be a subset of a topological space $X$ . A collection $U = {U_{α}}_{α \in A}$ of open sets is an open cover of $K$ if

K \subseteq α \in A ⋃ U_{α} .

⟳This is the defining inclusion — every point of

K

lies in at least one

U_{α}

A finite subcover is a finite subcollection ${U_{α_{1}}, \dots, U_{α_{n}}} \subseteq U$ that still covers:

K \subseteq i = 1 ⋃ n U_{α_{i}} .

⟳Restrict the index set

A

to a finite subset

{α_{1}, \dots, α_{n}}

while preserving the covering inclusion.

Here $X$ is the ambient space, $K$ the subset under study, $A$ an arbitrary (possibly uncountable) index set, and each $U_{α}$ an open set of $X$ .

Compactness

A subset $K \subseteq X$ is compact if every open cover of $K$ has a finite subcover. Symbolically,

\forall U open cover of K, \exists finite F \subseteq A : K \subseteq α \in F ⋃ U_{α} .

⟳Universally quantify over covers, existentially over finite subcollections — this is the literal reading of the definition.

Boundedness and Total Boundedness

A set $K$ in a metric space $(X, d)$ is bounded if it lies inside some ball:

\exists x_{0} \in X, R > 0 : K \subseteq B (x_{0}, R) = {x : d (x, x_{0}) < R} .

⟳One reference point and one finite radius enclose the whole set.

It is totally bounded if for every $ε > 0$ it can be covered by finitely many $ε$ -balls:

\forall ε > 0, \exists x_{1}, \dots, x_{m} \in X : K \subseteq i = 1 ⋃ m B (x_{i}, ε) .

⟳Total boundedness strengthens boundedness by demanding finite covers at every scale

ε

, not just one large ball.

Heine–Borel

Let $K \subseteq R^{n}$ with the standard Euclidean metric. The following are equivalent:

$K$ is compact (every open cover has a finite subcover);
$K$ is sequentially compact (every sequence in $K$ has a subsequence converging to a point of $K$ );
$K$ is closed and bounded;
$K$ is complete and totally bounded.

K \subseteq R^{n} compact ⟺ K closed and bounded .

⟳The headline equivalence (1) ⇔ (3) is the classical Heine–Borel statement; the other equivalences hold in any metric space and the dimensional finiteness of

R^{n}

is what links them to "closed and bounded."

The equivalence (3) ⇔ compactness is special to $R^{n}$ (and finite-dimensional normed spaces). In a general metric space, "closed and bounded" does not imply compact. The correct general equivalence is complete + totally bounded, statement (4).

Why It Holds

Closed and Bounded ⟹ Compact in ℝⁿ

We prove the harder direction of [6] for $R^{n}$ .

Step 1 — Reduce to a closed box. Since

K

is bounded,

K \subseteq Q_{0}

for some closed cube

Q_{0} = [- R, R]^{n}, R > 0.

⟳Boundedness [4] places

K

in a ball, which sits inside a cube of side

2 R

It suffices to show $Q_{0}$ is compact: a closed subset of a compact set is compact (a cover of $K$ together with the open set $R^{n} ∖ K$ covers $Q_{0}$ ; extract a finite subcover and discard $R^{n} ∖ K$ ).

Step 2 — Bisection / nested boxes. Suppose, for contradiction, that some open cover

U

Q_{0}

has no finite subcover. Bisect

Q_{0}

into

2^{n}

congruent subcubes by halving each side. If all

2^{n}

pieces admitted finite subcovers, their union

Q_{0}

would too. Hence at least one subcube

Q_{1}

has no finite subcover. Iterate to obtain nested cubes

Q_{0} \supseteq Q_{1} \supseteq Q_{2} \supseteq \dots, diam (Q_{k}) = \frac{diam ( Q _{0} )}{2 ^{k}} = \frac{2 R n}{2 ^{k}} .

⟳Each bisection halves the side length, hence halves the diameter; the diameter tends to

0

Step 3 — Nested interval / completeness. Each

Q_{k}

is a nonempty compact box, and the diameters shrink to

0

. By completeness of

R^{n}

(Cantor's nested-set property),

k = 0 ⋂ \infty Q_{k} = {p} for a unique point p \in Q_{0} .

⟳A decreasing sequence of nonempty closed sets with diameter

\to 0

in a complete space has a single common point.

Step 4 — Contradiction. Since

U

covers

Q_{0}

, some

U \in U

contains

p

; as

U

is open there is

δ > 0

with

B (p, δ) \subseteq U

. Choose

k

so large that

diam (Q_{k}) < δ

. Then

Q_{k} \subseteq B (p, δ) \subseteq U

, so

{U}

alone is a finite subcover of

Q_{k}

— contradicting that

Q_{k}

had no finite subcover. Hence

Q_{0}

(and therefore

K

) is compact.

■

Compact ⟹ Closed and Bounded — in any metric space

Bounded. Fix

x_{0} \in K

and cover

K

by the expanding balls

{B (x_{0}, m)}_{m \in N}

. A finite subcover has a largest radius

M

, so

K \subseteq B (x_{0}, M)

K

is bounded, satisfying [4]. Closed. Let

y \in / K

. For each

x \in K

, set

r_{x} = \frac{1}{2} d (x, y) > 0

; the balls

{B (x, r_{x})}_{x \in K}

cover

K

. Extract a finite subcover

B (x_{1}, r_{x_{1}}), \dots, B (x_{m}, r_{x_{m}})

, and let

r = min_{i} r_{x_{i}} > 0

. Then

B (y, r)

is disjoint from every

B (x_{i}, r_{x_{i}})

, hence disjoint from

K

. So

R^{n} ∖ K

is open and

K

is closed.

■

Émile Borel | 1871–1956

French mathematician who, in his 1894 thesis and 1895 paper, proved that a countable open cover of a closed bounded interval has a finite subcover — the first explicit "covering" statement. The result was later extended to arbitrary (uncountable) covers by Lebesgue, Schoenflies, and others. Borel went on to found modern measure theory (Borel sets, Borel measure) and made early contributions to probability and game theory. The "Heine" half of the name honors Eduard Heine, who in 1872 used the underlying compactness idea to prove that continuous functions on closed intervals are uniformly continuous.

Building Intuition

Think of an open cover as a fleet of overlapping "spotlights," each illuminating an open region. Compactness says: no matter how you arrange infinitely many spotlights to light up $K$ , you can switch off all but finitely many and still keep $K$ fully lit. A set fails compactness when it can "escape to the boundary or to infinity" — either by leaking out through a missing endpoint (not closed) or stretching off forever (not bounded). The bisection proof is a "divide and conquer": if no finite team of spotlights works, keep cutting the region in half, always keeping a troublesome piece; these pieces zoom in on a single point $p$ , but the one spotlight covering $p$ floods a whole neighborhood — large enough to swallow a tiny piece. Contradiction.

Bisection argument: nested boxes shrinking to the limit point p inside a covering open set U

Worked Examples

Testing compactness of three subsets of ℝ

Given. Decide whether each set is compact: (a)

A = [2, 5]

, (b)

B = (0, 1]

, (c)

C = {0} \cup {1/ n : n \in N}

. Steps.

(a) $A = [2, 5]$ is closed (contains both endpoints) and bounded (lies in $B (0, 6)$ ). By Heine–Borel [6], compact.
(b) $B = (0, 1]$ is bounded but not closed — the point $0$ is a limit point not in $B$ . Explicit escaping cover: $U = {(1/ n, 2) : n \in N}$ . Any finite subcollection has a largest $n = N$ , covering only $(1/ N, 2)$ , missing points near $0$ . No finite subcover ⟹ not compact.
(c) $C$ is bounded (in $[0, 1]$ ) and closed (its only limit point $0$ is included). Given any cover, one set $U_{0}$ contains $0$ , hence all but finitely many $1/ n$ ; cover the remaining finitely many points with one set each. Finite subcover exists.

Answer.

A compact, B not compact, C compact.

Extreme Value: bounding a continuous function via compactness

Given.

f (x, y) = x^{2} + 2 y^{2} - x

on the closed disk

K = {(x, y) : x^{2} + y^{2} \leq 1}

. Show

f

attains a maximum and find it. Steps.

$K$ is closed and bounded in $R^{2}$ , hence compact by [6]. Since $f$ is continuous, the Extreme Value Theorem guarantees a max and min are attained on $K$ .
Interior critical points: $\nabla f = (2 x - 1, 4 y) = (0, 0) \Rightarrow (x, y) = (\frac{1}{2}, 0)$ , inside $K$ . There $f = \frac{1}{4} - \frac{1}{2} = - \frac{1}{4}$ .
Boundary $x^{2} + y^{2} = 1$ : substitute $y^{2} = 1 - x^{2}$ , so $f = x^{2} + 2 (1 - x^{2}) - x = - x^{2} - x + 2$ for $x \in [- 1, 1]$ . Maximize $g (x) = - x^{2} - x + 2$ : $g^{'} (x) = - 2 x - 1 = 0 \Rightarrow x = - \frac{1}{2}$ , giving $g (- \frac{1}{2}) = - \frac{1}{4} + \frac{1}{2} + 2 = \frac{9}{4}$ . Check endpoints: $g (1) = 0, g (- 1) = 2$ .

Answer.

K max f = \frac{9}{4} at (- \frac{1}{2}, \pm \frac{3}{2}); K min f = - \frac{1}{4} at (\frac{1}{2}, 0) .

Compactness is exactly what licenses the assertion that these extrema exist before we compute them.

Common Misconceptions

"Closed and bounded means compact, always." False outside finite-dimensional spaces. The correct general characterization is complete + totally bounded [5]. In

R^{n}

closed ⇒ complete and bounded ⇒ totally bounded, which is why the simpler criterion works there — but this implication "bounded ⇒ totally bounded" fails in infinite dimensions. "Compactness = bounded." Boundedness alone is never enough:

(0, 1)

is bounded but not compact. And closedness alone is never enough:

[0, \infty)

is closed but not compact. You need both in

R^{n}

The closed unit ball in ℓ² is not compact

Let $ℓ^{2}$ be the space of square-summable real sequences with $∥ x ∥ = (\sum x_{i}^{2})^{1/2}$ , and let $K$ be its closed unit ball. Consider the standard basis vectors $e_{n} = (0, \dots, 0, 1, 0, \dots)$ (the $1$ in position $n$ ). Each $∥ e_{n} ∥ = 1$ , so $e_{n} \in K$ , and for $m \neq = n$ ,

∥ e_{m} - e_{n} ∥ = 1^{2} + 1^{2} = 2 .

⟳The two unit entries sit in different coordinates, so the difference has two nonzero components each equal to

1

No subsequence of $(e_{n})$ is Cauchy (all pairwise distances are $2$ ), so none converges. Thus $K$ is closed and bounded yet not sequentially compact, hence not compact. This shows Heine–Borel genuinely uses finite dimension. (In fact, Riesz's theorem: a normed space has compact closed unit ball iff it is finite-dimensional.)

Connections

Compactness threads through the whole of analysis. Via Sequential Compactness and the Bolzano–Weierstrass Theorem, it guarantees convergent subsequences. It underwrites the Extreme Value Theorem (continuous functions on compact sets attain extrema) and the Uniform Continuity Theorem (continuous on compact ⇒ uniformly continuous). The metric characterization "complete + totally bounded" links it to Completeness of Metric Spaces and Cauchy Sequences. Topologically it generalizes to Tychonoff's Theorem for products and interacts with Connectedness (continuous images of compact connected sets are compact and connected). The failure in infinite dimensions motivates Weak Topologies and Banach–Alaoglu Theorem, the right substitute in functional analysis.

Real-World Applications

Optimization — The existence half of nonlinear programming relies on minimizing continuous objectives over compact feasible sets (Weierstrass existence theorem); compactification (adding $+ \infty$ , coercivity arguments) is a standard trick when feasible sets are merely closed.
Numerical analysis — Convergence proofs for discretization schemes extract convergent subsequences of approximate solutions from compact sets of functions (Arzelà–Ascoli, a compactness theorem for $C (K)$ ).
Differential equations — The Peano existence theorem for ODEs and compactness methods in PDEs (Rellich–Kondrachov compact embeddings) produce solutions as limits of approximating sequences.
Economics / game theory — Existence of Nash equilibria (Kakutani fixed-point theorem) and of competitive equilibria requires compact, convex strategy/price simplices.
Computer graphics & geometry — Closed bounded (hence compact) solids guarantee that distance functions and ray–surface intersections attain minima, used in collision detection and rendering.
Control theory — Pontryagin's maximum principle and bang-bang controls use compactness of admissible control sets (often closed bounded subsets of $R^{n}$ ) to ensure optimal controls exist.
Machine learning — Compactness of parameter/hypothesis spaces underlies uniform convergence bounds and the existence of empirical risk minimizers.

Historical Development

1817 — Bernard Bolzano proves an early version of the least-upper-bound and intermediate-value properties, implicitly using boundedness arguments.
1872 — Eduard Heine, building on Dirichlet's lectures, uses the covering/uniform-continuity idea to show continuous functions on closed intervals are uniformly continuous; this is the conceptual seed of "Heine."
1872 — Karl Weierstrass and Heine rigorize the real numbers and the notion of limit points (Bolzano–Weierstrass property).
1894–1895 — Émile Borel proves, in his thesis and a published paper, that a countable open cover of $[a, b]$ admits a finite subcover.
1900–1904 — Pierre Cousin, Arthur Schoenflies, and Henri Lebesgue extend the result to arbitrary (uncountable) covers; Lebesgue's lectures popularize the modern statement.
1906 — Maurice Fréchet introduces abstract metric spaces and the term compact (initially meaning sequentially compact), framing the property abstractly.
1914–1924 — Felix Hausdorff and the Polish school formalize topology; the open-cover definition becomes standard, and "compact" is fixed to mean the finite-subcover property.
1918 — Frigyes Riesz proves that the closed unit ball is compact iff the normed space is finite-dimensional, pinpointing where Heine–Borel must fail.
1930 — Andrey Tychonoff proves arbitrary products of compact spaces are compact, the deepest general compactness theorem.

Subtleties & Limits

Compactness has subtle set-theoretic and infinite-dimensional dependencies that remain pedagogically and foundationally delicate:

Choice dependence. Tychonoff's theorem (arbitrary products of compacts are compact) is equivalent to the Axiom of Choice. Even the equivalence of "compact" and "sequentially compact" can fail in general topological spaces and depends on countable choice in some metric arguments. Determining the precise choice strength of various compactness statements is the domain of reverse mathematics; many such calibrations are ongoing research.
Infinite-dimensional substitutes. Since norm-compactness is too strong in function spaces, one replaces it with weak and weak-\* compactness (Banach–Alaoglu, Eberlein–Šmulian). Characterizing which sets are weakly compact, and how compactness interacts with reflexivity, drives modern functional analysis.
Quantitative compactness. Replacing the qualitative finite-subcover property with quantitative measures (entropy numbers, Kolmogorov widths, metric entropy of totally bounded sets) is central to approximation theory and statistical learning; sharp entropy estimates for many natural classes remain open.