The Compactness Strategy

From Local Information to a Finite Argument

The Subsequence Strategy used selected terms to test claims about an entire sequence. Compactness provides another way to control infinitely many objects: it can turn a collection of local statements into a finite collection that covers the set. This is useful when each point has its own neighborhood or its own estimate, but the conclusion requires one argument that works across the whole set.

The central compactness move is to start with an open cover and extract a finite subcover. In a metric space, compactness is also closely connected to subsequences: every sequence in a compact set has a convergent subsequence whose limit remains in the set. The equivalence lets us choose whichever form fits the proof. A cover is often best for a finite-selection argument; a sequence is often best for detecting failure of a uniform conclusion.

The finite subcover need not consist of sets of any special shape. Compactness says that some finite selection works, not that every finite selection works. Nor does it say that every open cover is finite. The point is that an infinite cover can be reduced to finitely many of its members while still covering \(K\).

Compactness and Subsequences in Metric Spaces

For metric spaces, the open-cover definition has a sequential counterpart. In the statement below, convergence is measured using the metric on \(X\), and the limit is required to belong to \(K\). That last condition matters: a sequence can converge in the surrounding space to a point outside the set.

Proof. First suppose \(K\) is compact, and take any sequence \((x_n)\) in \(K\). Call \(x\in K\) a cluster point of this sequence if every open ball centered at \(x\) contains \(x_n\) for infinitely many indices \(n\). There must be such a point. Otherwise, for every \(x\in K\), there would be a radius \(r_x>0\) such that the ball \(B(x,r_x)\) contains terms \(x_n\) for only finitely many indices. These balls cover \(K\). Compactness gives a finite subcover, so all terms of the sequence would have to occur among the finitely many indices associated with those finitely many balls. That is impossible, since the sequence has a term for every positive integer.

Choose a cluster point \(x\in K\). We can recursively choose increasing indices \(n_k\) so that the distance from \(x_{n_k}\) to \(x\) is less than \(1/k\). Indeed, every ball \(B(x,1/k)\) contains terms at infinitely many indices, so one can choose \(n_k\) larger than \(n_{k-1}\) and with \(x_{n_k}\in B(x,1/k)\). Then \(d(x_{n_k},x)<1/k\), and hence \(x_{n_k}\to x\). The limit belongs to \(K\), so \(K\) is sequentially compact.

Conversely, suppose \(K\) is sequentially compact. We first show that for every \(\varepsilon>0\), finitely many open balls of radius \(\varepsilon\), centered at points of \(K\), cover \(K\). If not, choose \(x_1\in K\), and then recursively choose \(x_{n+1}\in K\) outside the radius-\(\varepsilon\) balls centered at \(x_1,\ldots,x_n\). This gives \(d(x_i,x_j)\geq\varepsilon\) whenever \(i\neq j\). No subsequence can converge, since the terms of a convergent sequence must eventually be within \(\varepsilon/2\) of one another. This contradicts sequential compactness.

Now take any open cover \(\mathcal{U}\) of \(K\). There is a number \(\delta>0\) such that, for every \(x\in K\), the set \(B(x,\delta)\cap K\) lies in some member of \(\mathcal{U}\). To prove this, suppose no such \(\delta\) exists. For each positive integer \(n\), choose \(x_n\in K\) such that \(B(x_n,1/n)\cap K\) is not contained in any member of \(\mathcal{U}\). Sequential compactness gives a subsequence \(x_{n_k}\) converging to some \(x\in K\). Choose \(U\in\mathcal{U}\) containing \(x\). Since \(U\) is open, there is an \(r>0\) with \(B(x,r)\cap K\subseteq U\). For all sufficiently large \(k\), both \(d(x_{n_k},x)<r/2\) and \(1/n_k<r/2\). The triangle inequality then gives \(B(x_{n_k},1/n_k)\cap K\subseteq B(x,r)\cap K\subseteq U\), a contradiction.

Finally, cover \(K\) by finitely many balls of radius \(\delta/2\) centered at points \(p_1,\ldots,p_m\in K\), using the finite-ball property just proved. For each \(p_i\), choose \(U_i\in\mathcal{U}\) with \(B(p_i,\delta)\cap K\subseteq U_i\). Each set \(B(p_i,\delta/2)\cap K\) lies inside \(B(p_i,\delta)\cap K\subseteq U_i\). Thus \(U_1,\ldots,U_m\) cover \(K\), and \(\mathcal{U}\) has a finite subcover. Therefore \(K\) is compact. \(\square\)

The proof shows two useful mechanisms. Sequential compactness prevents points from staying separated by a fixed positive distance, which gives finite ball covers. It also prevents a sequence of points from repeatedly witnessing that no uniform neighborhood size works for a given open cover.

Compactness Turns Continuity into Uniform Continuity

Continuity at each point allows the required input tolerance to depend on the point. Uniform continuity asks for one tolerance that works at every point. Compactness is a way to obtain that uniform choice. The sequential form makes the contradiction particularly direct.

Proof. Suppose \(f\) is not uniformly continuous. Then there is an \(\varepsilon_0>0\) such that, for every positive integer \(n\), there are \(x_n,y_n\in K\) with \(d(x_n,y_n)<1/n\) but \(\rho(f(x_n),f(y_n))\geq\varepsilon_0\). By the Compactness and Sequential Compactness Theorem, some subsequence \(x_{n_k}\) converges to a point \(x\in K\). Since \(d(x_{n_k},y_{n_k})<1/n_k\) and \(1/n_k\to0\), the triangle inequality implies \(y_{n_k}\to x\) as well. Continuity at \(x\) gives \(f(x_{n_k})\to f(x)\) and \(f(y_{n_k})\to f(x)\). Another application of the triangle inequality yields

This contradicts \(\rho(f(x_{n_k}),f(y_{n_k}))\geq\varepsilon_0\) for every \(k\). Thus \(f\) is uniformly continuous. \(\square\)

Compactness Is Preserved by Continuous Maps

Another practical consequence is that a continuous map carries compact sets to compact sets. The proof follows the finite-subcover strategy exactly: pull the target cover back to the domain, use compactness there, and then transfer the finite selection back to the image.

Proof. Let \(\mathcal{V}\) be an open cover of \(f(K)\) by open sets in \(Y\). For each \(V\in\mathcal{V}\), continuity makes \(f^{-1}(V)\) open in \(K\). These inverse images cover \(K\), since every \(x\in K\) has \(f(x)\in f(K)\) and hence belongs to \(f^{-1}(V)\) for some \(V\in\mathcal{V}\). Compactness of \(K\) gives finitely many members \(V_1,\ldots,V_m\) whose inverse images cover \(K\). For every \(x\in K\), some \(f^{-1}(V_i)\) contains \(x\), so \(f(x)\in V_i\). Thus \(V_1,\ldots,V_m\) cover \(f(K)\), proving it is compact. \(\square\)

A Reliable Compactness Workflow

When compactness is available, first identify what must become finite or uniform. Then choose the compactness formulation that exposes that goal. Open covers are natural when the proof starts with local neighborhoods; subsequences are natural when the claim concerns limits or when failure can be witnessed by a sequence.

A common pitfall is to confuse compactness with boundedness or with the existence of a convergent subsequence whose limit might lie outside the set. Compactness in a metric space guarantees a subsequential limit in the set. Another pitfall is to claim a finite subcover without explaining how the finite selection covers the set. The open-cover definition requires both parts: the selection is finite, and its union still contains every point of \(K\).