Compactness — Heine-Borel and Bolzano-Weierstrass

⚡ Key Takeaways

Compactness is a topological property generalizing "closed and bounded." For subsets of $\mathbb{R}^n$ , the Heine–Borel theorem says compact = closed and bounded, while the Bolzano–Weierstrass theorem says every bounded sequence has a convergent subsequence. These results are the cornerstone of real analysis, guaranteeing maxima, uniform continuity, and the convergence machinery used throughout calculus, optimization, and differential equations. Named for Eduard Heine, Émile Borel, Bernard Bolzano, and Karl Weierstrass (1817–1895).

💡 The Core Insight

Compactness is finiteness in disguise. A compact set is one where any attempt to cover it with infinitely many open pieces always over-covers — finitely many already suffice. Equivalently, no sequence can "escape" — there is always a cluster point inside. This bridges the local (open sets, neighborhoods) and the global (the whole set), letting analysis trade unbounded processes for finite, manageable ones.

DefinitionOpen Cover and Compactness

Let $(X,d)$ be a metric space (or any topological space) and $K \subseteq X$ . A collection $\mathcal{U} = \{U_\alpha\}_{\alpha \in A}$ of open sets is an open cover of $K$ if

1

\quad K \subseteq \bigcup_{\alpha \in A} U_\alpha.

Derivation hint: this is the defining inclusion — every point of $K$ lies in at least one cover element. A finite subcover is a finite subcollection $U_{\alpha_1},\dots,U_{\alpha_n}$ with

2

\quad K \subseteq \bigcup_{i=1}^{n} U_{\alpha_i}.

Derivation hint: choose finitely many indices $\alpha_1,\dots,\alpha_n \in A$ that still cover $K$ . $K$ is compact if every open cover of $K$ has a finite subcover. Here $A$ is an arbitrary (possibly uncountable) index set.

DefinitionBoundedness, Closedness, Sequential Compactness

$K \subseteq \mathbb{R}^n$ is bounded if it fits inside some ball:

3

\quad \exists M > 0 : \quad \|x\| \le M \ \ \text{for all } x \in K.

Derivation hint: $M$ is any radius enclosing $K$ ; equivalently $\operatorname{diam}(K) < \infty$ . $K$ is closed if it contains all its limit points: whenever $x_k \in K$ and $x_k \to x$ , then $x \in K$ . $K$ is sequentially compact if every sequence $(x_k)_{k\ge1}$ in $K$ has a subsequence $(x_{k_j})$ converging to some limit $L \in K$ :

4

\quad \forall (x_k)\subseteq K\ \exists\, (x_{k_j}),\ L\in K:\quad \lim_{j\to\infty} x_{k_j}=L.

Derivation hint: extract indices $k_1<k_2<\cdots$ so the resulting subsequence has a limit inside $K$ .

TheoremHeine–Borel

A subset $K \subseteq \mathbb{R}^n$ is compact if and only if it is closed and bounded.

5

\quad K \text{ compact} \iff K \text{ closed} \ \wedge\ K \text{ bounded}.

Derivation hint: the forward direction holds in any metric space; the converse uses the completeness and finite-dimensionality of $\mathbb{R}^n$ .

TheoremBolzano–Weierstrass

Every bounded sequence in $\mathbb{R}^n$ has a convergent subsequence. Consequently, a subset $K\subseteq\mathbb{R}^n$ is compact iff it is sequentially compact. In a general metric space:

6

\quad K \text{ compact} \iff K \text{ sequentially compact} \iff K \text{ complete and totally bounded}.

Derivation hint: in metric spaces the three notions coincide; "totally bounded" replaces "bounded" once you leave $\mathbb{R}^n$ .

ProofBolzano–Weierstrass in

\mathbb{R}

via bisection

We show every bounded real sequence $(x_k)$ has a convergent subsequence.

Step 1. Since $(x_k)$ is bounded, all terms lie in an interval $I_0=[a_0,b_0]$ of length $L=b_0-a_0$ .

Step 2. Bisect $I_0$ into two halves. At least one half contains $x_k$ for infinitely many indices $k$ ; call it $I_1=[a_1,b_1]$ , of length $L/2$ .

Step 3. Repeat. Inductively obtain nested intervals

7

\quad I_0 \supseteq I_1 \supseteq I_2 \supseteq \cdots,\qquad |I_n| = \frac{L}{2^n},

Derivation hint: each bisection halves the length and keeps the half containing infinitely many terms. each containing infinitely many terms of the sequence.

Step 4. By the Nested Interval Property (a form of completeness of $\mathbb{R}$ ),

8

\quad \bigcap_{n=0}^{\infty} I_n = \{L^*\}

Derivation hint: lengths $\to 0$ and the intervals are nested, so the intersection is a single point $L^*$ . is a single point.

Step 5. Build the subsequence greedily: choose $k_1$ with $x_{k_1}\in I_1$ ; having chosen $k_j$ , pick $k_{j+1}>k_j$ with $x_{k_{j+1}}\in I_{j+1}$ (possible because $I_{j+1}$ holds infinitely many terms). Then $|x_{k_j}-L^*|\le |I_j| = L/2^j \to 0$ , so $x_{k_j}\to L^*$ . $\blacksquare$ For $\mathbb{R}^n$ , apply the result coordinate-by-coordinate: extract a subsequence convergent in coordinate 1, then a further subsequence convergent in coordinate 2, and so on through coordinate $n$ .

∎

ProofHeine–Borel, both directions

(⇒) Compact ⟹ closed and bounded. Bounded: The balls $\{B(0,m)\}_{m\in\mathbb{N}}$ form an open cover of $K$ . A finite subcover is contained in the largest ball $B(0,M)$ , so $K\subseteq B(0,M)$ is bounded. Closed: Suppose $x\notin K$ . For each $y\in K$ let $r_y=\tfrac12 d(x,y)>0$ . The balls $\{B(y,r_y)\}_{y\in K}$ cover $K$ ; take a finite subcover $B(y_1,r_{y_1}),\dots,B(y_m,r_{y_m})$ . Set $\delta=\min_i r_{y_i}>0$ . Then $B(x,\delta)$ misses all those balls, hence misses $K$ , so $x$ is not a limit point. Thus the complement of $K$ is open and $K$ is closed.

(⇐) Closed and bounded ⟹ compact (for $\mathbb{R}^n$ ).

Step 1 — the box is compact. First show a closed box $Q_0=[a,b]^n$ is compact, by contradiction. Suppose $\mathcal{U}$ is an open cover of $Q_0$ with no finite subcover. Bisect each edge to split $Q_0$ into $2^n$ sub-boxes. At least one sub-box $Q_1$ has no finite subcover. Iterate to get nested boxes

9

\quad Q_0 \supseteq Q_1 \supseteq \cdots, \qquad \operatorname{diam}(Q_k)=\frac{\operatorname{diam}(Q_0)}{2^k}\to 0,

Derivation hint: each step halves the diameter; "no finite subcover" passes to at least one child by pigeonhole. none admitting a finite subcover.

Step 2. By completeness, $\bigcap_k Q_k=\{p\}$ for a single point $p$ . Since $\mathcal{U}$ covers $Q_0$ , some $U\in\mathcal{U}$ contains $p$ , and being open, $U\supseteq B(p,\varepsilon)$ for some $\varepsilon>0$ .

Step 3. For $k$ large, $\operatorname{diam}(Q_k)<\varepsilon$ and $p\in Q_k$ , so $Q_k\subseteq B(p,\varepsilon)\subseteq U$ . Then $\{U\}$ alone is a finite subcover of $Q_k$ — contradicting its choice. Hence $Q_0$ is compact.

Step 4 — closed subset of compact is compact. Let $K$ be closed and bounded; then $K\subseteq Q_0$ for some box. Given an open cover $\mathcal{U}$ of $K$ , add the open set $X\setminus K$ to get an open cover of $Q_0$ . Extract a finite subcover; discarding $X\setminus K$ leaves finitely many elements of $\mathcal{U}$ covering $K$ . $\blacksquare$

∎

Figure

Bisection argument: nested boxes shrink onto a single point, trapped inside one cover element

U

KW

Karl Weierstrass

1815–1897

Weierstrass, the "father of modern analysis," made rigor the law of calculus. Working largely from lecture notes (he published little), he supplied the $\varepsilon$ – $\delta$ definitions that tamed limits, constructed the famous everywhere-continuous nowhere-differentiable function, and proved that a continuous function on a closed bounded interval attains its maximum and minimum — the Extreme Value Theorem, a direct child of compactness. The Bolzano–Weierstrass theorem bears his name because he rediscovered and disseminated, with full rigor, the cluster-point principle that Bolzano had stated decades earlier in relative obscurity.

🧠 Intuition

Think of compactness as "no room to run and no place to leak."

Bounded stops a sequence from running to infinity.
Closed stops a sequence from leaking out through a missing boundary point.

Remove either and points can escape. The open interval $(0,1)$ is bounded but not closed: the sequence $1/k$ tries to converge to $0$ , but $0$ has leaked out. The set $[0,\infty)$ is closed but not bounded: the sequence $1,2,3,\dots$ runs away. Only when both walls are up does every sequence get cornered into a cluster point that stays inside. The open-cover view says the same thing dually: if you try to tile a compact set with infinitely many small open patches, you always waste effort — finitely many already overlap to cover it. Non-compact sets defeat this: cover $(0,1)$ by $(1/n,\,1)$ for $n=2,3,\dots$ ; no finite subfamily reaches the points near $0$ .

ExampleThe closed unit ball in

\mathbb{R}^3

is compact

Given: $K=\{x\in\mathbb{R}^3 : \|x\|\le 1\}$ .

Step 1 — bounded: Every $x\in K$ satisfies $\|x\|\le 1$ , so $M=1$ works in equation [3]. ✔

Step 2 — closed: The map $x\mapsto\|x\|$ is continuous, and $K=\{x : \|x\|\le 1\}$ is the preimage of the closed set $[0,1]$ , hence closed. If $x_k\in K$ and $x_k\to x$ , then $\|x\|=\lim\|x_k\|\le 1$ , so $x\in K$ . ✔

Step 3 — apply Heine–Borel [5]: Closed and bounded in $\mathbb{R}^3 \Rightarrow$ compact.

\boxed{K=\overline{B}(0,1)\subseteq\mathbb{R}^3 \text{ is compact.}}

(Contrast: the open ball $\|x\|<1$ is bounded but not closed, hence not compact — the sequence $x_k=(1-1/k,0,0)$ converges to the boundary point $(1,0,0)\notin$ open ball.)

ExampleExtracting a convergent subsequence numerically

Given: the bounded sequence $x_k=\sin(k)$ , $k=1,2,3,\dots$ , all lying in $[-1,1]$ .

Goal: confirm Bolzano–Weierstrass produces a convergent subsequence and estimate a limit.

Step 1 — boundedness: $|x_k|=|\sin k|\le 1$ , so the sequence lives in the compact set $[-1,1]$ . By [6] a convergent subsequence exists.

Step 2 — bisection in action: Split $[-1,1]$ at $0$ . Computing terms: $\sin 1\approx0.841,\ \sin 2\approx0.909,\ \sin 3\approx0.141,\ \sin 6\approx-0.279,\ \sin 19\approx0.150,\ \sin 22\approx-0.0089,\ \sin 44\approx0.0177$ . Infinitely many terms accumulate near $0$ (whenever $k$ is close to a multiple of $\pi$ ).

Step 3 — identify a cluster point: Because $k \bmod \pi$ is equidistributed (Weyl), the values $\sin k$ are dense in $[-1,1]$ ; in particular some subsequence $x_{k_j}\to 0$ . Indices like $k=3,22,333,355,\dots$ (good rational approximations to multiples of $\pi$ ) give $\sin k_j$ ever closer to $0$ .

\boxed{\text{A subsequence } x_{k_j}=\sin(k_j)\to 0;\ \text{indeed every } L\in[-1,1]\text{ is a subsequential limit.}}

⚠️ Common mistake

"Closed and bounded always means compact." This is only true in finite-dimensional normed spaces. In an infinite-dimensional space (e.g. $\ell^2$ or $C[0,1]$ ) the closed unit ball is closed and bounded but not compact. The standard basis vectors $e_n=(0,\dots,0,1,0,\dots)$ satisfy $\|e_n\|=1$ and $\|e_n-e_m\|=\sqrt2$ for $n\ne m$ , so no subsequence is Cauchy — sequential compactness fails. Heine–Borel is a privilege of $\mathbb{R}^n$ .

⚠️ Common mistake

"Compactness is a property of the set alone." Compactness depends on the subspace topology but is intrinsic: a set is compact or not regardless of ambient space. However, closed and bounded are not intrinsic in the same way — $(0,1)$ is closed in itself yet not compact. Always check compactness via covers or sequences, not via "closed" applied to the wrong ambient space.

CounterexampleTotal boundedness is essential, not mere boundedness

In a general metric space, bounded ≠ totally bounded. Take an infinite set $X$ with the discrete metric $d(x,y)=1$ for $x\ne y$ . Then $X$ has diameter $1$ (bounded) and is closed and complete. But the cover by singleton balls $\{B(x,\tfrac12)\}=\{\{x\}\}$ has no finite subcover when $X$ is infinite. So $X$ is bounded and complete yet not compact — it fails to be totally bounded. This shows why [6] upgrades "bounded" to "totally bounded" outside $\mathbb{R}^n$ .

connection

Compactness is the engine behind several pillars of analysis. Via Continuity, it gives the Extreme Value Theorem (continuous image of a compact set is compact, hence attains max/min) and Heine–Cantor uniform continuity. It underlies Uniform Convergence through Dini's theorem and the Arzelà–Ascoli criterion, which characterizes compact sets of functions via equicontinuity. In Metric Spaces it equals "complete + totally bounded"; in general Topological Spaces it is defined purely by open covers, splitting from sequential compactness. It feeds Connectedness (both are "global" topological invariants preserved by continuous maps) and supports Completeness of the Real Numbers, since the Nested Interval Property used in the proofs is itself a completeness axiom.

🔬 Where this appears

⏳ History

- 1817 — Bernard Bolzano states and proves a version of the intermediate value theorem, implicitly using the least-upper-bound property and the cluster-point principle now called Bolzano–Weierstrass.

- 1865–1880s — Karl Weierstrass, in Berlin lectures, rigorously establishes that bounded sequences have convergent subsequences and that continuous functions on closed intervals attain extrema, popularizing

\varepsilon

–

\delta

rigor.

- 1872 — Eduard Heine publishes results on uniform continuity, using the covering idea for closed bounded intervals (the seed of Heine–Borel).

- 1894–1895 — Émile Borel proves that a countable open cover of a closed bounded interval has a finite subcover, in his thesis on measure and function theory.

- 1900s — Henri Lebesgue and Pierre Cousin extend Borel's covering theorem to arbitrary (uncountable) covers; the result becomes the modern Heine–Borel theorem.

- 1906 — Maurice Fréchet introduces metric spaces and the general notion of compactness (compacité), framing sequential compactness abstractly.

- 1914 — Felix Hausdorff systematizes topological spaces; the open-cover definition of compactness gains its general home.

- 1923–1929 — Pavel Alexandrov and Pavel Urysohn develop bicompactness (today's "compactness") and prove key theorems (e.g. compactness via filters/nets later by Cartan, Bourbaki).

- 1930 — Andrey Tychonoff proves arbitrary products of compact spaces are compact (Tychonoff's theorem), the deepest general result, equivalent to the Axiom of Choice.

open-problem

Several boundaries of compactness reveal deep structure rather than unsolved problems, but they remain conceptually subtle:

Compactness vs. sequential compactness diverge in general topological spaces. There exist compact spaces (e.g. $[0,1]^{[0,1]}$ , the product of continuum-many copies of $[0,1]$ ) that are not sequentially compact, and sequentially compact spaces that are not compact (e.g. the first uncountable ordinal $[0,\omega_1)$ ). Only for metric spaces do the two coincide.

Tychonoff's theorem and the Axiom of Choice. The statement "every product of compact spaces is compact" is equivalent to the Axiom of Choice (Kelley, 1950). Restricting to Hausdorff factors weakens it to the (strictly weaker) Boolean Prime Ideal theorem. Thus the strength of compactness in infinite products is entangled with the foundations of set theory.

Characterizing compact sets in function spaces is delicate and ongoing in applications: the Arzelà–Ascoli (equicontinuity) and Kolmogorov–Riesz (in $L^p$ ) criteria solve special cases, but identifying compact subsets of more exotic spaces drives current functional analysis.

Compactness — Heine-Borel and Bolzano-Weierstrass

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also