The Covering Strategy

From Countably Many Choices to Covering Arguments

The Diagonalization Strategy organized countably many requirements by making successive choices compatible. A covering argument addresses a different problem: local information may be given at every point, with no countable list of points or requirements available. An open cover lets us collect those local neighborhoods, and compactness reduces the collection to finitely many sets.

The finite subcover is more than a convenient simplification. In a compact metric space, an open cover also has a uniform scale: sufficiently small subsets of the space must fit inside a single member of the cover. This is the Lebesgue Number Lemma. It converts local neighborhoods, whose sizes may initially depend on the point, into one radius that works throughout the space.

Throughout, \(K\) is a subset of a metric space \((X,d)\), and open sets covering \(K\) are understood to be open in the relative topology on \(K\). Write

Thus \(B_K(x,r)\) is the open ball about \(x\) within \(K\), even if \(K\) is not itself open in \(X\).

Open Covers and a Uniform Scale

A family \(\mathcal U\) of subsets of \(K\) is an open cover of \(K\) if every \(U\in\mathcal U\) is relatively open and their union is \(K\). Compactness means that every open cover has a finite subcover. We use the Compactness and Sequential Compactness in Metric Spaces theorem from earlier in the course: for a compact metric space, every sequence has a convergent subsequence with limit in the space.

Proof. First suppose \(K=\varnothing\). There are no nonempty subsets \(A\subseteq K\), so the conclusion about such subsets holds vacuously; for example, take \(\delta=1\). Now assume \(K\neq\varnothing\).

We first prove the ball formulation. Suppose, to the contrary, that no positive radius works for every point of \(K\). Then, for each positive integer \(n\), there is a point \(x_n\in K\) such that \(B_K(x_n,1/n)\) is not contained in any member of \(\mathcal U\). By sequential compactness, some subsequence \((x_{n_j})\) converges to a point \(x\in K\). Since \(\mathcal U\) covers \(K\), choose \(U\in\mathcal U\) with \(x\in U\). The set \(U\) is relatively open, so there is an \(r>0\) such that \(B_K(x,r)\subseteq U\).

For all sufficiently large \(j\), convergence gives \(d(x_{n_j},x)<r/2\), and \(n_j\to\infty\) gives \(1/n_j<r/2\). If \(y\in B_K(x_{n_j},1/n_j)\), then the triangle inequality yields

Hence \(y\in B_K(x,r)\subseteq U\). We have shown \(B_K(x_{n_j},1/n_j)\subseteq U\), contradicting the choice of \(x_{n_j}\). Therefore some \(\delta>0\) works in the ball formulation.

Now let \(A\subseteq K\) be nonempty with \(\operatorname{diam}(A)<\delta\), and choose \(x\in A\). For each \(y\in A\), the definition of diameter gives \(d(x,y)\leq\operatorname{diam}(A)<\delta\). Thus \(A\subseteq B_K(x,\delta)\), which is contained in some member of \(\mathcal U\). This proves the subset formulation.

Conversely, if every nonempty subset of diameter less than \(\delta\) lies in a member of the cover, apply that statement to \(B_K(x,\delta/2)\). Any two points in this ball are at distance less than \(\delta\), so the ball has diameter at most \(\delta\), which is not quite enough to use the stated strict inequality. Instead apply the subset statement to \(B_K(x,\delta/3)\): its diameter is at most \(2\delta/3<\delta\). This gives a ball around each \(x\) contained in a cover member, with radius \(\delta/3\). Thus the two formulations express the same uniform-scale principle, allowing a change in the numerical radius. \(\square\)

The compactness hypothesis enters at a precise point: it supplies a convergent subsequence of points that would otherwise witness failure at smaller and smaller scales. Openness then provides a neighborhood of the subsequential limit, and convergence forces a sufficiently small ball around a sufficiently late point into that neighborhood.

Compact Spaces Have Finite Nets

A second useful covering conclusion is that compact metric spaces can be approximated by finitely many points at any prescribed positive accuracy. The finite collection of points need not belong to a particular set in an open cover; its role is to provide a finite set of centers that comes within a chosen distance of every point.

Proof. If \(K=\varnothing\), the empty set is a finite \(\varepsilon\)-net for every \(\varepsilon>0\), because there are no points \(x\in K\) to check. Now suppose \(K\neq\varnothing\), and fix \(\varepsilon>0\). The family of relative open balls \(\{B_K(x,\varepsilon):x\in K\}\) covers \(K\), since \(x\in B_K(x,\varepsilon)\) for every \(x\in K\). By compactness, finitely many of these balls cover \(K\): there are \(a_1,\ldots,a_m\in K\) such that

Set \(F=\{a_1,\ldots,a_m\}\). For every \(x\in K\), the finite-cover inclusion puts \(x\) in some \(B_K(a_i,\varepsilon)\), which means \(d(x,a_i)<\varepsilon\). Thus \(F\) is a finite \(\varepsilon\)-net. Since \(\varepsilon>0\) was arbitrary, \(K\) is totally bounded. \(\square\)

How to Use a Covering Argument

The Lebesgue Number Lemma is especially useful when a problem gives local information in the form “for each point, there is a neighborhood where a property holds.” Those neighborhoods form an open cover. Compactness first produces finitely many neighborhoods; the Lebesgue number then identifies a scale at which small sets cannot cross between unrelated local descriptions.

For instance, suppose a compact interval is covered by open sets on each of which a particular local estimate is valid. Once a Lebesgue number is known, any subinterval shorter than that number is contained in one member of the cover, so the estimate can be applied throughout that subinterval. This is a common way to pass from local control to a finite argument. The exact conclusion still depends on what the local estimate says; the covering result supplies the organization and scale, not the estimate itself.

A frequent error is to select one neighborhood at each point and assume their radii have a positive lower bound. Pointwise openness gives a radius that can depend on the point; it does not by itself give a common radius. Compactness is what makes the uniform conclusion possible. The empty-space case also deserves care: a finite subcover may be empty, so an argument that takes the minimum of its radii is valid only after the nonempty case has been established.

The covering strategy begins with local sets and asks what compactness can make finite or uniform. Finite subcovers reduce the number of cases; Lebesgue numbers control the size of subsets that fit within a single case; finite nets replace an entire compact space by finitely many nearby centers. Before using any of these conclusions, verify the cover, the topology, the compactness hypothesis, and whether the empty case needs separate treatment.