Compact Spaces and Open Covers

Compactness is the load-bearing finiteness hypothesis across mathematics and its applications. In real and functional analysis, it underlies the extraction of convergent subsequences (via Sequential Compactness) used in existence proofs for differential equations and in the calculus of variations. In differential geometry, compact manifolds have finite volume, finite fundamental groups in favorable cases, and support globally-defined extremal problems (geodesics realizing minimal distance, curvature bounds via Bonnet–Myers). In complex analysis, compactness of closed bounded regions underlies the maximum modulus principle's practical use and Montel's theorem on normal families. In algebraic topology and algebraic geometry, compact (or "proper") spaces allow intersection numbers and degrees to be well-defined finite quantities rather than divergent sums. In optimization and control theory, compactness of feasible regions is precisely the hypothesis that guarantees a continuous cost function attains a minimum — the abstract content of the corollary proved above. In computer science and logic, compactness of certain topological spaces of models (Stone spaces) underlies the model-theoretic compactness theorem for first-order logic, and finite-subcover arguments appear directly in proofs of correctness for covering-based algorithms. In physics, compact configuration or momentum spaces (such as compactified extra dimensions in string theory) rely on exactly the same open-cover finiteness to ensure well-posed path integrals and spectral decompositions.

Ask the tutor: Why does the Heine–Borel theorem fail for the closed unit ball in an infinite-dimensional normed space?

Think of an open cover as a collection of "flashlights," each illuminating some open region of the space, with every point of the space lit by at least one flashlight. Compactness says you can always find a finite set of flashlights that still lights up the whole space, no matter how the cover was chosen. This is a strong, non-obvious constraint: it must hold for every open cover, including deliberately adversarial ones designed to need infinitely many pieces. The classic non-example is the open interval $(0,1)$. Cover it by the sets $U_n = (1/n, 1)$ for $n = 2, 3, 4, \dots$. Each point of $(0,1)$ lies in some $U_n$, so this is a genuine open cover, but no finite subcollection covers $(0,1)$: a finite union of these sets is just the largest one, $U_N = (1/N, 1)$, which misses the points near $0$. This single cover is the reason $(0,1)$ fails to be compact — it is "open at one end," and the open cover can chase that missing endpoint forever. Contrast this with $[0,1]$: no matter how you cover it by open sets, the compactness of $[0,1]$ (proved via the Heine-Borel Theorem) guarantees a finite subcover always exists, precisely because $[0,1]$ contains both of its endpoints and has no "missing point" for an open cover to approach without ever reaching.

Example 1: A compactness argument with an infinite, asymmetric open cover. Let $X = \{0\} \cup \{1/n : n \in \mathbb{Z}_{\ge 1}\} \subset \mathbb{R}$ with the subspace topology from $\mathbb{R}$ (this space is often called a "convergent sequence with its limit"). We verify directly, from the open-cover definition, that $X$ is compact — without invoking Heine–Borel, to illustrate the definition in action. Let $\{U_\alpha\}_{\alpha \in A}$ be an arbitrary open cover of $X$ (open in $\mathbb{R}$, or equivalently in $X$; it does not matter by the Lemma). Since $0 \in X$, some member of the cover contains $0$; call it $U_{\alpha_0}$. Because $U_{\alpha_0}$ is open in $\mathbb{R}$ and contains $0$, there is $\varepsilon > 0$ with $(-\varepsilon, \varepsilon) \subseteq U_{\alpha_0}$. Choose $N$ so large that $1/N < \varepsilon$; then every point $1/n$ with $n \ge N$ lies in $(-\varepsilon,\varepsilon) \subseteq U_{\alpha_0}$. So $U_{\alpha_0}$ alone already covers $0$ and all but finitely many points of $X$, namely it may miss only $1, 1/2, \dots, 1/(N-1)$. For each of these finitely many remaining points $1/k$ ($k = 1, \dots, N-1$), choose some $U_{\alpha_k}$ from the original cover containing it (such a set exists because $\{U_\alpha\}$ covers all of $X$). Then

is a finite subcover. Since $\{U_\alpha\}$ was an arbitrary open cover, $X$ is compact. Notice precisely where the limit point $0 \in X$ was used: it is exactly the presence of $0$ in the space that let one open set $U_{\alpha_0}$ absorb infinitely many of the $1/n$'s at once. If we deleted $0$ from $X$, leaving $Y = \{1/n : n \ge 1\}$, the same cover-construction as in the intuition section (using intervals around each $1/n$ that avoid all other points) would show $Y$ is not compact — a discrete infinite set is never compact, since the cover by singletons $\{1/n\}$ (each open in the subspace topology on $Y$, since points of $Y$ are isolated) has no finite subcover.

Example 2: Building a genuinely non-compact set and pinpointing the failing cover. Let $Z = [0,1) \cup \{2\} \subset \mathbb{R}$, again with the subspace topology. We show $Z$ is not compact by exhibiting one explicit open cover with no finite subcover — and we check carefully that the cover really is open in $Z$ and really has no finite subcover, since sloppy verification here is the most common student error. Define, for $n = 2, 3, 4, \dots$,

each an open subset of $\mathbb{R}$ (a union of two open intervals), hence $U_n \cap Z$ is open in $Z$. Note $U_n \cap Z = [0, 1-\tfrac1n) \cup \{2\}$. Cover verification. Given any $x \in Z$: if $x = 2$, then $x \in U_n$ for every $n$. If $x \in [0,1)$, then since $x < 1$ we can choose $n$ large enough that $1 - 1/n > x$ (possible because $1 - 1/n \to 1$ as $n \to \infty$), so $x \in U_n$. Hence $\{U_n\}_{n \ge 2}$ covers $Z$. No finite subcover. Suppose, for contradiction, that finitely many sets $U_{n_1}, \dots, U_{n_k}$ (with $n_1 < \cdots < n_k$) covered $Z$. Since the intervals $[0, 1-1/n)$ are nested increasing in $n$, the union $U_{n_1} \cup \cdots \cup U_{n_k}$ equals $U_{n_k}$, whose intersection with $Z$ is $[0, 1 - 1/n_k) \cup \{2\}$. But this misses every point $x \in Z$ with $1 - 1/n_k \le x < 1$ — and such points exist, e.g. $x = 1 - \tfrac{1}{2n_k}$, which satisfies $1 - 1/n_k < x < 1$ so $x \in Z$ but $x \notin U_{n_k}\cap Z$. This contradicts the assumption that $U_{n_1}, \dots, U_{n_k}$ cover $Z$. Hence no finite subcollection of $\{U_n\}$ covers $Z$, so $Z$ is not compact. Consistently with part (a) of the theorem: $Z$ is not closed in $\mathbb{R}$ (it is missing the limit point $1$), so this is not a counterexample to any theorem — it simply shows that "closed" really is doing work, matching the Heine–Borel picture that compact subsets of $\mathbb{R}$ must be closed.