In a general [topological space](/page/Topology), the topology can be extraordinarily large --- the collection of open sets may have cardinality far exceeding that of the underlying set. This abundance creates serious obstacles. When we attempt to extract convergent subsequences, construct countable approximations, or reduce open covers to manageable subcovers, we need the topology itself to be "small" in a precise sense. The question is: *what condition on a topology ensures that the entire open-set structure is governed by a countable amount of data?* The answer is **second countability**: the existence of a countable base for the topology. A base is a collection of open sets from which every open set can be reconstructed by taking unions, so a countable base means that every open set in the space --- no matter how complicated --- is a union of sets drawn from a single countable family. This is a far-reaching constraint. It forces the space to be [separable](/page/Separable%20Space) (admit a countable dense subset), [Lindelof](/page/Lindel%C3%B6f%20Space) (every open cover has a countable subcover), and --- when combined with regularity --- [metrizable](/page/Metrizable%20Space) (by the Urysohn Metrization Theorem). In the metrizable setting, second countability, separability, and the Lindelof property are all equivalent, but in general topology they diverge, and second countability is the strongest of the three. The importance of second countability is most visible in two areas. First, in the theory of [metrizable spaces](/page/Metrizable%20Space), second countability is the countability axiom that appears in the Urysohn Metrization Theorem --- the first and most widely used sufficient condition for metrizability. A topological space that is regular and second-countable embeds into the Hilbert cube $[0,1]^{\mathbb{N}}$, and is therefore metrizable. Second, in the definition of [Polish spaces](/page/Polish%20Space), which form the natural setting for descriptive set theory and probability, a Polish space is by definition a separable, completely metrizable space --- and in the metrizable setting, separability is equivalent to second countability. The entire apparatus of Borel complexity theory, regular conditional distributions, and the Kuratowski isomorphism theorem rests on this countability condition. [example: The Sorgenfrey Line --- Separable but Not Second-Countable] The distinction between separability and second countability is most sharply illustrated by the **Sorgenfrey line** $\mathbb{R}_\ell$: the real line equipped with the lower limit topology, whose basic open sets are the half-open intervals $[a, b)$ for $a < b$. **Separability.** The rationals $\mathbb{Q}$ form a countable dense subset: for any $x \in \mathbb{R}$ and $\varepsilon > 0$, the basic open set $[x, x + \varepsilon)$ contains a rational number (by density of $\mathbb{Q}$ in $\mathbb{R}$). Therefore $\mathbb{R}_\ell$ is separable. **Failure of second countability.** Suppose for contradiction that $\mathcal{B} = \{B_1, B_2, \ldots\}$ is a countable base for $\mathbb{R}_\ell$. For each $x \in \mathbb{R}$, the set $[x, x+1)$ is open, so there exists $B_{n(x)} \in \mathcal{B}$ with $x \in B_{n(x)} \subset [x, x+1)$. In particular, $\inf B_{n(x)} = x$ (since $B_{n(x)} \subset [x, x+1)$ forces $\inf B_{n(x)} \ge x$, while $x \in B_{n(x)}$ gives $\inf B_{n(x)} \le x$). If $x \neq y$, then $\inf B_{n(x)} \neq \inf B_{n(y)}$, so $B_{n(x)} \neq B_{n(y)}$, whence $n(x) \neq n(y)$. The map $x \mapsto n(x)$ is an injection from $\mathbb{R}$ (uncountable) into $\mathbb{N}$ (countable), which is impossible. This example shows that a space can have a countable dense subset (controlling the "point-level complexity") while its topology is too rich to be generated by any countable family of open sets (its "open-set complexity" is uncountable). Second countability is strictly stronger than separability. [/example] ## Definition The Sorgenfrey line example exposes a gap between two levels of "countable control." Having a countable dense subset tells us that the *points* of the space are accessible via countable approximation, but it says nothing about the *open sets* --- the topology may require uncountably many "building blocks" that no countable collection can replace. To close this gap, we need a condition that bounds the complexity of the open-set structure itself, not just the point set. The right condition turns out to be simple: require that a countable subcollection of open sets suffices to generate the entire topology via unions. Recall that a **base** (or **basis**) for a topology $\tau$ on a set $X$ is a subcollection $\mathcal{B} \subset \tau$ such that every open set $G \in \tau$ can be written as a union of members of $\mathcal{B}$. Equivalently, $\mathcal{B}$ is a base if and only if for every $G \in \tau$ and every $x \in G$, there exists $B \in \mathcal{B}$ with $x \in B \subset G$. [definition: Second-Countable Space] A [topological space](/page/Topology) $(X, \tau)$ is **second-countable** (also called **second axiom of countability**) if it admits a countable base for its topology. That is, there exists a countable collection $\mathcal{B} = \{B_1, B_2, \ldots\} \subset \tau$ such that every open set $G \in \tau$ is a union of members of $\mathcal{B}$: \begin{align*} G = \bigcup \{B \in \mathcal{B} : B \subset G\}. \end{align*} [/definition] The "second" in "second-countable" distinguishes this property from **first countability**, which requires only that each point $x \in X$ have a countable neighbourhood base --- a countable collection $\{U_1, U_2, \ldots\}$ of open sets containing $x$ such that every open set containing $x$ includes some $U_n$. Every second-countable space is first-countable (restrict the countable base to those members containing $x$), but the converse fails: the Sorgenfrey line is first-countable (the sets $\{[x, x + 1/n)\}_{n=1}^\infty$ form a countable neighbourhood base at $x$) but not second-countable, as shown above. Second countability depends on the topology, not on the underlying set. The same set $\mathbb{R}$ is second-countable in the Euclidean topology (with base $\{(p, q) : p, q \in \mathbb{Q}, \, p < q\}$) and not second-countable in the lower limit topology (as demonstrated above) or the discrete topology (where any base must contain all singletons, which is uncountable). [example: Second Countability of $\mathbb{R}^n$] Euclidean space $\mathbb{R}^n$ with the standard topology is second-countable. A countable base is the collection of open balls with rational centres and rational radii: \begin{align*} \mathcal{B} = \{B(q, r) : q \in \mathbb{Q}^n, \, r \in \mathbb{Q}, \, r > 0\}. \end{align*} This is countable as a subset of $\mathbb{Q}^n \times \mathbb{Q}_{>0}$, which is a countable product of countable sets. To verify it is a base: let $G \subset \mathbb{R}^n$ be open and $x \in G$. There exists $\varepsilon > 0$ with $B(x, \varepsilon) \subset G$. Choose $q \in \mathbb{Q}^n$ with $|x - q| < \varepsilon/3$ and $r \in \mathbb{Q}$ with $\varepsilon/3 < r < \varepsilon/2$. Then $x \in B(q, r)$ (since $|x - q| < \varepsilon/3 < r$) and $B(q, r) \subset B(x, \varepsilon) \subset G$ (since for $y \in B(q, r)$, $|y - x| \le |y - q| + |q - x| < r + \varepsilon/3 < \varepsilon/2 + \varepsilon/3 < \varepsilon$). [/example] ## Separability and the Lindelof Property Having defined second countability as a condition on the open sets, we face two immediate questions: does controlling the open-set structure force a countable dense subset to exist? And does it force open covers to admit countable subcovers? Both answers are affirmative, and neither requires any additional hypotheses beyond second countability itself. ### From Second Countability to Separability The first consequence is that every second-countable space is [separable](/page/Separable%20Space). If a topology can be generated by countably many open sets, then countably many "sample points" --- one from each basis element --- already approximate the entire space. [quotetheorem:1067] The construction is straightforward: let $\mathcal{B} = \{B_1, B_2, \ldots\}$ be a countable base for $\tau$. For each nonempty $B_n$, choose a point $d_n \in B_n$ (by the Axiom of Choice, or more precisely by countable choice). The set $D = \{d_n : B_n \neq \varnothing\}$ is countable. To verify density: let $G \in \tau$ be nonempty. Then $G$ contains some $B_n$ (since $\mathcal{B}$ is a base), so $d_n \in D \cap G$. Since every nonempty open set meets $D$, we have $\overline{D} = X$. The converse fails in general topological spaces --- the Sorgenfrey line is the canonical counterexample (separable but not second-countable). The construction above produces a dense subset but gives no control over its algebraic or topological structure: the set $D$ is not in general a subgroup, a subring, or a sublattice of $X$, even when $X$ carries such structure. The converse does hold in metrizable spaces, a fact of central importance discussed in the section on equivalence in metrizable spaces below. ### From Second Countability to the Lindelof Property The second consequence is more surprising and more powerful: every second-countable space is **Lindelof** --- every open cover admits a countable subcover. This is a strong property that sits between compactness (finite subcovers) and the trivial observation that every cover is a subcover of itself. [definition: Lindelof Space] A topological space $(X, \tau)$ is **Lindelof** if every open cover of $X$ has a countable subcover. That is, whenever $\{U_\alpha\}_{\alpha \in A}$ is a collection of open sets with $X = \bigcup_{\alpha \in A} U_\alpha$, there exist countably many indices $\alpha_1, \alpha_2, \ldots \in A$ such that $X = \bigcup_{k=1}^\infty U_{\alpha_k}$. [/definition] [quotetheorem:1104] The argument uses the countable base to reduce any open cover to a countable one. Let $\mathcal{B} = \{B_1, B_2, \ldots\}$ be a countable base and let $\{U_\alpha\}_{\alpha \in A}$ be an open cover of $X$. For each $x \in X$, choose $\alpha(x) \in A$ with $x \in U_{\alpha(x)}$, and then (since $\mathcal{B}$ is a base) choose $B_{n(x)} \in \mathcal{B}$ with $x \in B_{n(x)} \subset U_{\alpha(x)}$. The collection $\{B_{n(x)} : x \in X\}$ is a subcollection of the countable family $\mathcal{B}$, so it uses at most countably many distinct basis elements. For each basis element $B_m$ that appears, fix a single choice of $\alpha$ with $B_m \subset U_\alpha$, and call it $\alpha_m$. The countable subcollection $\{U_{\alpha_m}\}$ covers $X$: for any $x \in X$, the basis element $B_{n(x)}$ is one of those selected, so $x \in B_{n(x)} \subset U_{\alpha_{n(x)}}$. The theorem does not assert that the countable subcover has any particular structure --- it may involve redundant or overlapping sets, and there is no guarantee that the subcover is efficient or that its elements are drawn from any particular subcollection. The strength of the result lies entirely in the existence of *some* countable subcover. The Lindelof property is strictly weaker than compactness and strictly stronger than being a countable union of compact sets. The real line $\mathbb{R}$ is Lindelof (it is second-countable) but not compact. The Sorgenfrey line is Lindelof (a separate, more involved argument) but not second-countable, showing that the Lindelof property does not imply second countability in general. [example: The Lindelof Property in Action --- $\sigma$-Compactness of $\mathbb{R}^n$] The Lindelof property of $\mathbb{R}^n$ immediately implies that $\mathbb{R}^n$ is $\sigma$-compact: it can be written as a countable union of compact sets. Consider the open cover $\{B(0, k)\}_{k=1}^\infty$. This is already countable, so it trivially has a countable subcover. But more powerfully, consider *any* open cover $\{U_\alpha\}$ of $\mathbb{R}^n$. The Lindelof property extracts a countable subcover $\{U_{\alpha_k}\}_{k=1}^\infty$. Since $\overline{B}(0, m) \cap \bigcup_{k=1}^{N(m)} U_{\alpha_k} \supset \overline{B}(0, m)$ for $N(m)$ large enough (by compactness of $\overline{B}(0, m)$), every compact subset of $\mathbb{R}^n$ is covered by finitely many elements of the countable subcover. This interplay between the Lindelof property and local compactness is used throughout measure theory: the Lindelof property ensures that locally finite measures on $\mathbb{R}^n$ are $\sigma$-finite, because a locally finite measure assigns finite mass to each compact set, and the Lindelof property ensures we need only countably many such sets to cover the space. [/example] ## Equivalence in Metrizable Spaces The implications established above --- second-countable implies separable, second-countable implies Lindelof --- are strict in general topological spaces. The Sorgenfrey line shows that the converses fail: it is both separable and Lindelof, yet not second-countable. A natural question arises: *under what additional structural hypothesis do these three conditions become equivalent, so that verifying any one of them automatically yields the other two?* The answer is metrizability. In the presence of a metric, the hierarchy \begin{align*} \text{second-countable} \implies \text{separable}, \quad \text{second-countable} \implies \text{Lindelof} \end{align*} collapses into a single equivalence. This is one of the most frequently invoked results in the interplay between topology and analysis, and it explains why, in the metrizable setting, one can freely substitute "separable" for "second-countable" without loss. [quotetheorem:1068] Each implication in this equivalence has a distinct character, and understanding the mechanism behind each clarifies why the equivalence requires metrizability. [explanation: The Three Implications] **$(1) \Rightarrow (2)$:** This holds in all topological spaces, as established above. No metric is needed. **$(1) \Rightarrow (3)$:** This also holds in all topological spaces. No metric is needed. **$(2) \Rightarrow (1)$:** This is where the metric is essential. Given a countable dense subset $D = \{d_1, d_2, \ldots\}$, construct the countable collection \begin{align*} \mathcal{B} = \{B(d_k, 1/m) : k, m \in \mathbb{N}\}. \end{align*} This is countable (indexed by $\mathbb{N} \times \mathbb{N}$). To verify it is a base: let $G \subset X$ be open and $x \in G$. There exists $r > 0$ with $B(x, r) \subset G$. By density of $D$, choose $d_k \in D$ with $d(x, d_k) < r/3$. Choose $m \in \mathbb{N}$ with $1/m \in (r/3, 2r/3)$ (such $m$ exists by the Archimedean property). Then $x \in B(d_k, 1/m)$ since $d(x, d_k) < r/3 < 1/m$. For any $y \in B(d_k, 1/m)$, the triangle inequality gives $d(y, x) \le d(y, d_k) + d(d_k, x) < 1/m + r/3 < 2r/3 + r/3 = r$, so $B(d_k, 1/m) \subset B(x, r) \subset G$. The metric enters twice: first to produce the explicit basis elements $B(d_k, 1/m)$, and second to verify the containment via the triangle inequality. In a general topological space, there is no way to "inflate" the countable dense subset $D$ into a countable base --- the topology may require uncountably many open sets that cannot be built from balls around the points of $D$. **$(3) \Rightarrow (2)$:** This also uses the metric. For each $m \in \mathbb{N}$, the collection $\{B(x, 1/m) : x \in X\}$ is an open cover of $X$. By the Lindelof property, extract a countable subcover $\{B(x_{m,k}, 1/m) : k \in \mathbb{N}\}$. The set $D = \{x_{m,k} : m, k \in \mathbb{N}\}$ is countable (a countable union of countable sets). For any $x \in X$ and $\varepsilon > 0$, choose $m$ with $1/m < \varepsilon$. Since $\{B(x_{m,k}, 1/m)\}_k$ covers $X$, there exists $k$ with $x \in B(x_{m,k}, 1/m)$, giving $d(x, x_{m,k}) < 1/m < \varepsilon$. So $D$ is dense. [/explanation] The failure of $(2) \Rightarrow (1)$ in general topological spaces is the reason that, in non-metrizable settings, one must carefully distinguish between separability and second countability. In [functional analysis](/page/Functional%20Analysis), where weak and weak* topologies are rarely metrizable on the full space, this distinction is critical: a [Banach space](/page/Banach%20Space) $X$ is always separable in its norm topology if and only if it is second-countable in its norm topology (since the norm topology is metrizable), but the weak topology $\sigma(X, X^*)$ may be separable without being second-countable. [example: A Separable Topology That Is Not Second-Countable --- the Weak Topology on $\ell^1$] The Banach space $\ell^1$ is separable in its norm topology (the sequences with finitely many nonzero rational entries form a countable dense subset). By the Schur property of $\ell^1$, weak convergence and norm convergence coincide for sequences, so one might expect the weak topology to behave similarly to the norm topology. However, the weak topology $\sigma(\ell^1, \ell^\infty)$ on $\ell^1$ is **not** first-countable (by the same Baire category argument used for the weak topology on any infinite-dimensional Banach space --- see the [Metrizable Space](/page/Metrizable%20Space) page), and therefore not second-countable. Yet $\ell^1$ is weakly separable: the same countable dense subset that works for the norm topology (finite rational sequences) is also weakly dense, because every norm-open set is weakly open (the weak topology is coarser than the norm topology), so norm density implies weak density. This gives a separable topology that is not second-countable. [/example] ## Hereditary Properties A natural question about any topological property is whether it passes to subspaces, products, and quotients. For second countability, the answer is particularly clean for subspaces and countable products, but fails for uncountable products and quotients. Understanding these permanence properties is essential for establishing second countability of new spaces constructed from known ones. ### Subspaces Second countability passes to all subspaces --- not just open or closed ones, but arbitrary subspaces. This makes second countability a **hereditary** property, a feature that separability conspicuously lacks in general topological spaces. [quotetheorem:1069] The proof is immediate: if $\mathcal{B} = \{B_1, B_2, \ldots\}$ is a countable base for $\tau$, then $\mathcal{B}_A = \{B_n \cap A : n \in \mathbb{N}\}$ is a countable base for the subspace topology on $A$. (Every open set in $A$ has the form $G \cap A$ for some $G \in \tau$; since $G = \bigcup \{B_n : B_n \subset G\}$, we get $G \cap A = \bigcup \{B_n \cap A : B_n \subset G\}$.) This hereditary property has two important consequences. First, since every second-countable space is separable and every subspace of a second-countable space is second-countable, every subspace of a second-countable space is also separable. This recovers the fact that subspaces of separable metrizable spaces are separable --- a result that fails in general topological spaces (a subspace of a separable space need not be separable in the absence of second countability). Second, hereditary second countability means that when proving second countability of a subspace, it suffices to establish it for the ambient space. In particular, every subset of $\mathbb{R}^n$ (with the subspace topology) is second-countable, and every subspace of a [Polish space](/page/Polish%20Space) is second-countable (though it need not be Polish --- it may fail to be completely metrizable). [example: Separability of Subspaces Fails Without Second Countability] To see why hereditary second countability is stronger than hereditary separability, consider the Sorgenfrey plane $\mathbb{R}_\ell \times \mathbb{R}_\ell$ --- the product of two copies of the Sorgenfrey line. The Sorgenfrey plane is separable: $\mathbb{Q} \times \mathbb{Q}$ is a countable dense subset (for any $(x, y) \in \mathbb{R}^2$ and $\varepsilon > 0$, the basic open set $[x, x+\varepsilon) \times [y, y+\varepsilon)$ contains a point of $\mathbb{Q} \times \mathbb{Q}$). However, the anti-diagonal $A = \{(x, -x) : x \in \mathbb{R}\}$ is a subspace that is **not** separable (in the subspace topology inherited from $\mathbb{R}_\ell \times \mathbb{R}_\ell$). The subspace topology on $A$ is the discrete topology: for each $x \in \mathbb{R}$, the set $[x, x+1) \times [-x, -x+1)$ is open in $\mathbb{R}_\ell \times \mathbb{R}_\ell$ and intersects $A$ in the single point $\{(x, -x)\}$, so every singleton in $A$ is open. A discrete uncountable space is not separable (every dense subset must contain every point). This failure is impossible in a second-countable space, since second countability is hereditary and implies separability. The Sorgenfrey plane, being separable but not second-countable, allows this pathology. [/example] ### Countable Products Second countability is preserved under countable products with the [product topology](/page/Product%20Topology). This is the key to establishing second countability for infinite-dimensional spaces like $\mathbb{R}^{\mathbb{N}}$ (the space of all real sequences) and $\{0, 1\}^{\mathbb{N}}$ (the Cantor space). [quotetheorem:1070] The construction of a countable base for the product proceeds as follows. For each $k$, let $\mathcal{B}_k = \{B_{k,1}, B_{k,2}, \ldots\}$ be a countable base for $\tau_k$. The product topology has a base consisting of all finite intersections of sets of the form $\pi_k^{-1}(B_{k,j})$, where $\pi_k$ is the $k$-th coordinate projection. More explicitly, a basis for the product topology consists of all sets of the form \begin{align*} B_{k_1, j_1} \times B_{k_2, j_2} \times \cdots \times B_{k_N, j_N} \times \prod_{k \notin \{k_1, \ldots, k_N\}} X_k, \end{align*} where $N \in \mathbb{N}$, $k_1 < \cdots < k_N$, and $j_1, \ldots, j_N \in \mathbb{N}$. This collection is countable: it is indexed by finite subsets of $\mathbb{N} \times \mathbb{N}$ (the pairs $(k_i, j_i)$), and the collection of all finite subsets of a countable set is countable. The restriction to countable products is essential. Uncountable products of second-countable spaces with more than one point are never second-countable. The product $\{0, 1\}^{\mathbb{R}}$ is not second-countable (and not even separable), despite each factor being finite. [remark: Uncountable Products Destroy Second Countability] If $(X, \tau)$ is a topological space with at least two points, then $X^A$ (with the product topology) is second-countable only if $A$ is countable. To see this, suppose $X^A$ has a countable base $\mathcal{B}$. Choose distinct points $a, b \in X$ and disjoint open sets $U, V \subset X$ with $a \in U$ and $b \in V$. Define the point $p \in X^A$ by $p(\alpha) = a$ for all $\alpha$. For each $\alpha \in A$, the set $W_\alpha := \pi_\alpha^{-1}(U)$ is an open neighbourhood of $p$, so there exists $B_{n(\alpha)} \in \mathcal{B}$ with $p \in B_{n(\alpha)} \subset W_\alpha$. We claim $n(\alpha) \neq n(\beta)$ whenever $\alpha \neq \beta$: suppose $n(\alpha) = n(\beta)$. Then $B_{n(\alpha)} \subset W_\alpha \cap W_\beta = \pi_\alpha^{-1}(U) \cap \pi_\beta^{-1}(U)$. But consider the point $q$ defined by $q(\gamma) = a$ for $\gamma \neq \beta$ and $q(\beta) = b$. Then $q \in W_\alpha$ (since $q(\alpha) = a \in U$) but $q \notin W_\beta$ (since $q(\beta) = b \notin U$). Since $B_{n(\alpha)} \subset W_\beta$, we have $q \notin B_{n(\alpha)}$. On the other hand, consider the point $q'$ defined by $q'(\gamma) = a$ for $\gamma \neq \alpha$ and $q'(\alpha) = b$. Then $q' \notin W_\alpha$, so $q' \notin B_{n(\alpha)}$. The basis element $B_{n(\alpha)}$ must distinguish $\alpha$ from $\beta$, so it must "know about" both coordinates. Since any basis element in the product topology restricts only finitely many coordinates, each $B_{n(\alpha)}$ can constrain at most finitely many coordinates --- but it must constrain $\alpha$ specifically (to stay inside $W_\alpha$). Different coordinates $\alpha$ require different basis elements. Thus the map $\alpha \mapsto n(\alpha)$ is injective from $A$ into $\mathbb{N}$, forcing $A$ to be countable. [/remark] ### Quotients and Continuous Images Second countability is preserved under continuous open surjections (in particular, open quotient maps). If $f: X \to Y$ is a continuous, open, surjective map and $X$ is second-countable, then $\{f(B_n) : n \in \mathbb{N}\}$ is a countable base for $Y$ (openness ensures $f(B_n)$ is open; the base property follows from surjectivity). However, second countability is **not** preserved under arbitrary continuous surjections or arbitrary quotient maps (the quotient map may create "too many" new open sets or destroy the countable base structure). ## The Role in Metrization Second countability appears as the crucial hypothesis in the two most important metrization theorems. Without it, a topological space may satisfy every separation axiom and still fail to be metrizable. With it, only regularity is needed. ### The Urysohn Metrization Theorem The problem of metrization is to determine when a topological space $(X, \tau)$, defined purely in terms of open sets, admits a compatible metric --- a function $d: X \times X \to [0, \infty)$ satisfying the metric axioms whose induced topology equals $\tau$. The topology provides no distance function, no notion of "how far apart" two points are, and the challenge is to construct such a function from the abstract open-set data. The Urysohn Metrization Theorem resolves this problem for second-countable spaces by showing that regularity is the only additional requirement. [quotetheorem:992] The "only if" direction is straightforward: every metrizable space is regular (disjoint open sets separate points from closed sets, using balls of appropriate radii), and if the metrizable space is also second-countable, both conditions are satisfied. The content is in the "if" direction: given only regularity and a countable base $\mathcal{B}$, one constructs an embedding into $[0, 1]^{\mathbb{N}}$. The idea is to enumerate all pairs $(B_n, B_m)$ of basis elements with $\overline{B_n} \subset B_m$. By regularity, there are "enough" such pairs to separate points from closed sets. For each such pair, [Urysohn's Lemma](/page/Urysohn's%20Lemma) (applicable because regular second-countable spaces are normal) provides a continuous function $f_{nm}: X \to [0, 1]$ with $f_{nm}|_{\overline{B_n}} \equiv 0$ and $f_{nm}|_{X \setminus B_m} \equiv 1$. The countably many functions $\{f_{nm}\}$ are assembled into a single map $\Phi: X \to [0, 1]^{\mathbb{N}}$, which is a topological embedding. The embedding into the Hilbert cube has a powerful consequence: every second-countable regular space is homeomorphic to a subspace of a compact metrizable space. In particular, it is separable (since subspaces of separable metrizable spaces are separable) and has at most continuum-many open sets (since $[0, 1]^{\mathbb{N}}$ has at most $2^{\aleph_0}$ open sets). This theorem is the principal reason that second countability is included in the standard definition of a smooth manifold. A smooth manifold is required to be Hausdorff (which, combined with the locally Euclidean structure, gives regularity) and second-countable. The Urysohn Metrization Theorem then guarantees that every smooth manifold is metrizable --- a fact essential for partition-of-unity arguments, Riemannian geometry, and integration on manifolds. ### The Nagata-Smirnov Metrization Theorem The Urysohn theorem requires a countable base, which excludes metrizable spaces that are "too large" to be second-countable --- for instance, an uncountable discrete space. The Nagata-Smirnov Metrization Theorem provides a complete characterisation of metrizability by replacing the countable base with a weaker condition. [quotetheorem:995] A **$\sigma$-locally finite base** is a base $\mathcal{B} = \bigcup_{n=1}^\infty \mathcal{B}_n$ where each $\mathcal{B}_n$ is locally finite: every point has a neighbourhood meeting only finitely many members of $\mathcal{B}_n$. Every countable base is $\sigma$-locally finite (each $\mathcal{B}_n = \{B_n\}$ is trivially locally finite), so the Urysohn theorem is a corollary. The Nagata-Smirnov theorem handles the general case, including uncountable metrizable spaces. For the purposes of analysis, the Urysohn theorem suffices for the vast majority of applications, since the spaces encountered in analysis (function spaces, Sobolev spaces, spaces of distributions) are typically separable, hence second-countable in their metrizable topologies. ## Countable Products and the Hilbert Cube A natural question arises from the permanence results: is there a single "universal" second-countable space into which every second-countable space embeds? Such a universal space would serve as a concrete ambient space for the entire theory, much as $\mathbb{R}^n$ serves as an ambient space for smooth manifolds. The countable product construction provides the answer: the **Hilbert cube** $[0, 1]^{\mathbb{N}} = \prod_{k=1}^\infty [0, 1]$, equipped with the product topology. The Hilbert cube is second-countable (as a countable product of second-countable spaces), compact (by [Tychonoff's theorem](/page/Tychonoff's%20Theorem), as a product of compact spaces), and metrizable (as a countable product of metrizable spaces, with the compatible metric $d(x, y) = \sum_{k=1}^\infty 2^{-k} |x_k - y_k|$). It is therefore a compact, second-countable, metrizable space. The Urysohn Metrization Theorem shows that the Hilbert cube is a **universal space** for the class of second-countable regular spaces: every such space embeds into $[0, 1]^{\mathbb{N}}$. This universality makes the Hilbert cube the natural "ambient space" for second-countable topology, analogous to the role of $\mathbb{R}^n$ as an ambient space for smooth manifolds. [example: Universality of the Hilbert Cube] The following spaces all embed into the Hilbert cube. **$\mathbb{R}^n$.** Define $\varphi: \mathbb{R} \to (0, 1)$ by $\varphi(t) = \arctan(t)/\pi + 1/2$. This map is continuous and strictly increasing, with continuous inverse $\varphi^{-1}(s) = \tan(\pi(s - 1/2))$, so $\varphi$ is a homeomorphism from $\mathbb{R}$ onto $(0, 1)$. The product map $\Phi: \mathbb{R}^n \to [0,1]^{\mathbb{N}}$ defined by $\Phi(x_1, \ldots, x_n) = (\varphi(x_1), \ldots, \varphi(x_n), 0, 0, \ldots)$ is a topological embedding. **The Cantor space $\{0, 1\}^{\mathbb{N}}$.** Since $\{0, 1\} \subset [0, 1]$, the inclusion $\{0, 1\}^{\mathbb{N}} \hookrightarrow [0, 1]^{\mathbb{N}}$ is a topological embedding (the product topology on $\{0, 1\}^{\mathbb{N}}$ agrees with the subspace topology inherited from $[0, 1]^{\mathbb{N}}$). **The Baire space $\mathbb{N}^{\mathbb{N}}$.** Define $\Psi: \mathbb{N}^{\mathbb{N}} \to [0,1]^{\mathbb{N}}$ by $\Psi(n_1, n_2, \ldots) = (1/(n_1 + 1), 1/(n_2 + 1), \ldots)$. Each coordinate map $n_k \mapsto 1/(n_k + 1)$ is a homeomorphism from $\mathbb{N}$ (discrete) onto $\{1/(m+1) : m \in \mathbb{N}\} \subset [0, 1]$ (also discrete as a subspace of $[0, 1]$), so $\Psi$ is a topological embedding by the universal property of the product. **Every separable Banach space.** Any separable Banach space $X$ is second-countable (since its norm topology is metrizable) and regular. By the Urysohn Metrization Theorem, $X$ embeds into $[0, 1]^{\mathbb{N}}$. That such diverse spaces --- some compact, some locally compact, some neither; some complete, some not --- all embed into a single compact metrizable space illustrates the remarkable range of the Hilbert cube. [/example] ## Failure and Boundaries of Second Countability Many spaces that arise naturally in analysis and topology fail to be second-countable, and these failures have concrete consequences: metrization theorems do not apply, hereditary separability breaks down, and the Lindelof property may be lost. Identifying the precise mechanism of failure --- whether the space is "too large" (uncountable discrete spaces), "too rich" (non-separable Banach spaces), or "too weak" (weak topologies) --- clarifies which tools remain available and which must be replaced. ### The Discrete Topology on Uncountable Sets The simplest failure: if $X$ is uncountable, the discrete topology on $X$ is metrizable (with the discrete metric $d(x, y) = 1$ for $x \neq y$) but not second-countable. Any base for the discrete topology must contain every singleton $\{x\}$ (since $\{x\}$ is open and cannot be written as a union of other open sets unless they all contain $x$, in which case they must equal $\{x\}$). This gives a base of cardinality $|X|$, which is uncountable. This is the most basic example of a metrizable space that fails second countability, and it shows that metrizability alone does not imply second countability --- the Nagata-Smirnov theorem applies to the discrete space (the base $\{\{x\} : x \in X\}$ is $\sigma$-locally finite with $\mathcal{B}_1 = \{\{x\} : x \in X\}$ locally finite), while the Urysohn theorem does not. ### Non-Separable Banach Spaces In functional analysis, the distinction between separable and non-separable spaces is the distinction between second-countable and non-second-countable (in the norm topology, which is metrizable). The space $\ell^\infty$ of bounded real sequences is the canonical non-separable Banach space: the uncountable family $\{\mathbb{1}_S : S \subset \mathbb{N}\}$ of indicator sequences satisfies $\|\mathbb{1}_S - \mathbb{1}_T\|_{\ell^\infty} = 1$ for $S \neq T$, so no countable subset is dense, and $\ell^\infty$ is not second-countable in its norm topology. The non-second-countability of $\ell^\infty$ has concrete analytical consequences: the weak* topology on $(\ell^\infty)^* = (\ell^1)^{**}$ restricted to the unit ball is **not** metrizable (in contrast to the weak* topology on $B_{X^*}$ when $X$ is separable, which is metrizable). This means that the Banach-Alaoglu theorem provides weak* compactness of the dual unit ball of $\ell^\infty$, but not sequential weak* compactness --- one must work with nets rather than sequences. ### Weak Topologies on Infinite-Dimensional Spaces A particularly instructive class of failures arises from the weak topology $\sigma(X, X^*)$ on an infinite-dimensional [Banach space](/page/Banach%20Space) $X$. Even when $X$ is separable (so that its norm topology is second-countable), the weak topology is **not** second-countable --- indeed, it is not even first-countable. The argument, detailed on the [Metrizable Space](/page/Metrizable%20Space) page, uses the Baire Category Theorem: a countable neighbourhood base at the origin would force $X^*$ to be a countable union of finite-dimensional subspaces, contradicting the Baire property of the complete metric space $X^*$. Since first countability is necessary for second countability, the weak topology on an infinite-dimensional Banach space is never second-countable. This failure is precisely why the Banach-Alaoglu theorem guarantees only net-convergence (not sequential convergence) in the weak topology of a non-reflexive space, and why separate arguments (such as the Eberlein-Smulian theorem) are needed to establish sequential weak compactness. ## Standard Techniques for Verifying Second Countability In practice, the challenge is not stating the definition of second countability but *verifying* it for a given space. Constructing a countable base directly can be tedious and is often the wrong approach --- it is frequently easier to establish second countability indirectly, by exploiting the equivalence with separability (in the metrizable case), by embedding into a known second-countable space, or by reducing to permanence properties. The following techniques cover the most common situations. ### Direct Construction of a Countable Base The most elementary method: explicitly exhibit a countable collection of open sets and verify the base axiom. This is the approach used for $\mathbb{R}^n$ (rational-centre, rational-radius balls) and works whenever the space has a natural family of "basic" open sets parametrised by countable data. **General pattern.** Identify a family of open sets $\{B(d, r)\}$ parametrised by a countable set $D$ (typically a dense subset) and countable radii $R$ (typically $\{1/m : m \in \mathbb{N}\}$ or $\mathbb{Q}_{>0}$). Verify the base property using the triangle inequality or an analogous approximation argument. **When to use.** When the space is a concrete metric space and a countable dense subset is known. ### The Separability-Metrizability Bridge In a metrizable space, separability and second countability are equivalent. This means that to verify second countability, it suffices to find a countable dense subset --- often a much easier task than constructing a countable base directly. **General pattern.** Establish that the space is metrizable (by exhibiting a compatible metric or applying a metrization theorem). Then find a countable dense subset, using any of the standard techniques: rational approximation, density of polynomials, density of smooth functions, or density of finite-rank operators. **When to use.** When the space is known to be metrizable and a density result is available. [example: Second Countability of $C(K)$ for Compact Metrizable $K$] Let $K$ be a compact metrizable space. The space $C(K)$ of continuous real-valued functions on $K$, equipped with the supremum norm, is a Banach space and hence metrizable. To establish second countability, it suffices to show separability. Since $K$ is compact and metrizable, $K$ is second-countable. By the Stone-Weierstrass theorem, the continuous functions on $K$ that are generated by a countable family separating points are dense. More concretely, when $K = [0, 1]$, the polynomials with rational coefficients $\{\sum_{k=0}^N q_k x^k : N \in \mathbb{N}_0, \, q_k \in \mathbb{Q}\}$ form a countable dense subset by the Weierstrass Approximation Theorem. Since $C([0, 1])$ is separable and metrizable, it is second-countable. [/example] ### Embedding into a Second-Countable Space To prove that $X$ is second-countable, it suffices to embed $X$ (homeomorphically) into a space already known to be second-countable. This follows from the hereditary property of second countability. **General pattern.** Construct a continuous injection $\Phi: X \hookrightarrow Y$ that is a homeomorphism onto its image, where $Y$ is second-countable. Then $\Phi(X)$ is second-countable (as a subspace of $Y$), and $X \cong \Phi(X)$ is second-countable. **When to use.** When the space naturally embeds into a product of second-countable spaces or into the Hilbert cube. ### Product and Subspace Reductions To verify second countability of a space defined as a subspace or countable product, appeal to the permanence properties: - **Subspaces:** Any subspace of a second-countable space is second-countable. - **Countable products:** Any countable product of second-countable spaces is second-countable. These reductions are used constantly. For example, the [Sobolev space](/page/Sobolev%20Spaces) $W^{k,p}(U)$ (for $1 \le p < \infty$ and $U \subset \mathbb{R}^n$ open) embeds isometrically into a finite product of $L^p(U)$ spaces (one copy for each multi-index $|\alpha| \le k$). Since $L^p(U)$ is separable and metrizable (hence second-countable), the finite product is second-countable, and $W^{k,p}(U)$ --- as a subspace --- is second-countable. ## References 1. Munkres, J. R., *Topology* (2000). 2. Kelley, J. L., *General Topology* (1955). 3. Willard, S., *General Topology* (1970). 4. Engelking, R., *General Topology* (1989). 5. Brezis, H., *Functional Analysis, Sobolev Spaces and Partial Differential Equations* (2011). 6. Kechris, A. S., *Classical Descriptive Set Theory* (1995).