From a Sigma-Discrete Base to a Metric
A metrization proof must do more than produce many open sets: it must organize them at successively finer scales so that distinct points can be distinguished and neighborhoods can be recovered from the resulting metric. Bing’s metrization theorem makes a sigma-discrete base sufficient for that organization, provided the space has the appropriate separation property. The proof has two stages. First, the base is used to build covers whose stars become small and whose successive levels refine one another. Then a standard cover-to-metric construction turns those covers into a compatible metric.
Recall that a family \(\mathcal{F}\) of subsets of \(X\) is discrete if every point has an open neighborhood that meets at most one member of \(\mathcal{F}\). A base \(\mathcal{B}\) is sigma-discrete if it is a countable union \(\mathcal{B}=\bigcup_{n\in\mathbb{N}}\mathcal{B}_n\), where each \(\mathcal{B}_n\) is a discrete family. A space is regular here in the \(T_1\) sense: points are closed, and whenever a point \(x\) lies in an open set \(O\), there is an open set \(V\) with \(x\in V\subseteq\overline{V}\subseteq O\).
The \(T_1\) part of regularity matters. Refinement conditions alone do not guarantee that distinct points are separated. The covers used in the proof must become fine around each point, and the separation hypothesis ensures that this local control can distinguish points rather than merely reproduce a topology in which different points have identical neighborhoods.
Stars and Controlled Covers
A star-refinement controls not just each small cover element, but also all elements that meet it. The useful output of Bing’s base construction is a sequence of open covers \((\mathcal{U}_n)_{n\in\mathbb{N}}\) with two properties: \(\mathcal{U}_{n+1}\) star-refines \(\mathcal{U}_n\); and, for every \(x\in X\) and every open neighborhood \(O\) of \(x\), some \(n\) satisfies \(\operatorname{St}(\{x\},\mathcal{U}_n)\subseteq O\). The second condition says that the stars around \(x\) form a local system fine enough to fit inside any prescribed neighborhood.
Here is the role of the sigma-discrete base in outline. For a requested open cover, regularity lets one choose smaller open neighborhoods whose closures stay inside the requested sets. Base elements can then be chosen inside those neighborhoods. The base is processed in its discrete layers: within one layer, neighborhoods can be arranged so that a point has local contact with at most one member. This permits the choices for that layer to be made without uncontrolled overlap. Repeating the refinement at successively finer stages, while retaining the earlier cover as a target, gives the star-refining sequence. The technical refinement lemma packages this induction; it is the main construction step in Bing’s proof.
Discreteness is stronger than local finiteness. Local finiteness allows a neighborhood to meet finitely many family members, perhaps several; discreteness requires at most one. That distinction is valuable when refining: one can control a layer member-by-member without needing to coordinate arbitrarily many intersecting members near a point. A sigma-locally-finite base supports a related metrization argument, but the particular advantage here is the sharper one-member-at-a-time control of each discrete layer.
Worked Example: A Shrinking Sequence of Covers of the Real Line
Let \(\mathbb{R}\) have its standard topology. For \(n\in\mathbb{N}\), put \(r_n=4^{-n}\), and let \(\mathcal{U}_n=\{(a-r_n,a+r_n):a\in\mathbb{R}\}\). Each \(\mathcal{U}_n\) is an open cover, since \(x\in(x-r_n,x+r_n)\) for every \(x\in\mathbb{R}\).
Fix \(U=(a-r_n,a+r_n)\). If \(V=(b-r_n,b+r_n)\) meets \(U\), then \(|a-b|<2r_n\). Every point of \(V\) is therefore at distance less than \(3r_n\) from \(a\): for \(y\in V\), \(|y-a|\leq |y-b|+|b-a|<r_n+2r_n=3r_n\). Since \(r_{n-1}=4r_n\) for \(n\geq2\), all such \(V\) lie in \((a-r_{n-1},a+r_{n-1})\), a member of \(\mathcal{U}_{n-1}\). Thus \(\mathcal{U}_n\) star-refines \(\mathcal{U}_{n-1}\).
For a fixed \(x\), a member \((a-r_n,a+r_n)\) containing \(x\) has \(|a-x|<r_n\). Every point in that member is less than \(2r_n\) from \(x\). Hence \(\operatorname{St}(\{x\},\mathcal{U}_n)\subseteq(x-2r_n,x+2r_n)\). Given an open neighborhood \(O\) of \(x\), choose \(\delta>0\) with \((x-\delta,x+\delta)\subseteq O\); since \(2\cdot4^{-n}\) tends to zero, some \(n\) has \(2r_n<\delta\). This verifies the local star condition. The calculation illustrates the cover conditions, not the sigma-discrete-base construction itself.
From Covers to a Compatible Metric
The standard cover-metrization lemma states the needed bridge: if a \(T_1\) space has a sequence of open covers in which each cover star-refines its predecessor and the point-stars eventually lie inside every neighborhood, then the topology is metrizable. Both hypotheses are essential to the intended conclusion. Star-refinement supplies consistency across scales; the local star condition ensures that the cover scales detect the given topology. The \(T_1\) hypothesis prevents distinct points from remaining indistinguishable at every scale.
One way to understand the lemma is through chains. At a chosen scale, points lying in one cover member are treated as close. A chain is a finite sequence of such steps, where the steps may use different scales; assign finer-scale steps smaller lengths and take the infimum of the lengths of chains joining two points. The nested star-refinement condition controls how chains at fine scales sit inside coarser cover elements. This control is what makes the chain distance satisfy the triangle inequality and makes small metric neighborhoods agree with the original open neighborhoods. The full chain-metric verification is the standard cover-metrization argument; it is not the part specific to Bing’s criterion.
The separation condition also has a direct cover interpretation. If \(x\ne y\), then \(X\setminus\{y\}\) is an open neighborhood of \(x\), because \(X\) is \(T_1\). Choose \(n\) with \(\operatorname{St}(\{x\},\mathcal{U}_n)\subseteq X\setminus\{y\}\). No member of \(\mathcal{U}_n\) can contain both \(x\) and \(y\), since such a member would put \(y\) in that star. Thus the cover scales distinguish the two points, as a metric must.
Proof sketch of Bing’s criterion. Assume \(X\) is regular and has a sigma-discrete base. Apply the discrete-layer refinement construction successively to obtain open covers \(\mathcal{U}_n\) such that \(\mathcal{U}_{n+1}\) star-refines \(\mathcal{U}_n\), and such that for every \(x\) and every open neighborhood \(O\) of \(x\), some point-star is contained in \(O\). Regularity supplies the nested neighborhoods required by the refinement construction; discreteness controls overlaps within each layer; countably many layers allow the refinements to be organized into a sequence. Since regularity includes \(T_1\), the cover-metrization lemma applies and gives a metric inducing exactly the original topology. This proves the criterion, with the technical work concentrated in the stated refinement and cover-metrization steps. \(\square\)
Worked Example: A Finite Discrete Space
Let \(X=\{p,q,r\}\) with the discrete topology \(\mathcal{T}=\mathcal{P}(X)\). The family \(\mathcal{B}_1=\{\{p\},\{q\},\{r\}\}\) is a base and is discrete: each point has its singleton open neighborhood, which meets exactly one member of the family. Thus this is a sigma-discrete base with just one nonempty layer.
Take \(\mathcal{U}_n=\{\{p\},\{q\},\{r\}\}\) for every \(n\). The star of any singleton member is that singleton, so each cover star-refines itself. Also \(\operatorname{St}(\{p\},\mathcal{U}_n)=\{p\}\), and the corresponding equalities hold for \(q\) and \(r\). These stars fit inside every open neighborhood of their points. The compatible metric can be written \(d(x,y)=0\) when \(x=y\), and \(d(x,y)=1\) when \(x\ne y\). Its open ball of radius \(1/2\) about \(x\) is \(\{x\}\), so it induces exactly the discrete topology.
Check: the metric separates distinct points because \(d(p,q)=d(p,r)=d(q,r)=1>0\); its triangle inequality holds since every nonzero distance is \(1\), while a sum of two distances joining distinct endpoints is at least \(1\).
A Common Pitfall: Local Finiteness Is Not Discreteness
Worked Example: Two Locally Finite Intervals That Are Not Discrete
In \(\mathbb{R}\), consider \(\mathcal{F}=\{(-1,1),(0,2)\}\). This family is locally finite: every point has any open neighborhood, and that neighborhood meets at most the two members of \(\mathcal{F}\). But the family is not discrete. The point \(1\) belongs to \((0,2)\) but not to \((-1,1)\). Every open neighborhood of \(1\) contains points less than \(1\) and sufficiently close to \(1\), which belong to \((-1,1)\); it also contains points greater than \(1\) and sufficiently close to \(1\), which belong to \((0,2)\). Therefore every neighborhood of \(1\) meets both members.
Common wrong argument: “Each point lies in only finitely many members, so the family is discrete.” That conclusion confuses membership with neighborhood intersection. In this example each point belongs to at most two intervals, yet neighborhoods of \(1\) meet both intervals. Local finiteness allows finitely many such intersections; discreteness demands a neighborhood meeting at most one. Bing’s layer-by-layer control uses the stronger property.
The central idea is thus a sequence of controlled scales, not simply a large supply of basis elements. A sigma-discrete base makes it possible to refine covers with especially strong overlap control. Regularity makes the refinements fit inside prescribed neighborhoods, and the \(T_1\) condition ensures that distinct points can be separated. Once those ingredients yield star-refining covers whose point-stars become arbitrarily small, the standard cover-to-metric construction completes the metrization.
Check Your Understanding
Use the definitions and cover conditions above to check the role of each hypothesis.
- Explain why the star of a singleton need not itself be a member of the cover.
- In the real-line example, why does \(|a-b|<2r_n\) hold when the two radius-\(r_n\) intervals intersect?
- Which part of the cover conditions ensures that the covers get fine relative to the original topology?
- Why does \(T_1\) imply that \(X\setminus\{y\}\) is an open neighborhood of \(x\) when \(x\ne y\)?
- Give the distinction between local finiteness and discreteness using the two intervals in the pitfall example.