Tutorials › Topology (Undergraduate) › Constructing Countably Many Separating Functions

Metrization Theorems · Tutorial 385 of 799

Constructing Countably Many Separating Functions

Learn how nested base pairs produce a countable family of continuous functions that separates points and detects the topology.

Advanced 12 min read

What You'll Learn

  • Identify the base pairs needed to construct separating functions.
  • See why a regular second-countable space is normal.
  • Apply Urysohn's lemma to each suitable pair of closed sets.
  • Prove the resulting countable family separates distinct points.
  • Recover each open set from the functions in the family.
  • Check the construction in the real line and in explicit countable spaces.

From a Countable Base to Function Coordinates

Continuous functions can record topological information about points. To distinguish two points, it is enough to find a continuous function that takes different values at them. A stronger goal is to choose functions that also recover every open set: around each point of an open set, one function should be positive there and vanish outside the open set. In a regular second-countable space, a countable base supplies the data for making these choices countably many times.

Definition: A space is second-countable if its topology has a countable base \(\mathcal{B}\): every open set is a union of members of \(\mathcal{B}\). We use the convention that regular means \(T_1\) and that a point and a closed set not containing it have disjoint open neighborhoods. A space is normal if any two disjoint closed sets have disjoint open neighborhoods.

The key base pairs are not arbitrary. We need pairs \(B,C\in\mathcal{B}\) for which \(\overline{B}\subseteq C\). For such a pair, the closed sets \(\overline{B}\) and \(X\setminus C\) are disjoint. Urysohn's lemma can then give a continuous function equal to \(1\) on \(\overline{B}\) and \(0\) on \(X\setminus C\). We first justify that the required normality is available.

Why Regularity and a Countable Base Give Normality

Proposition: Every regular second-countable space is normal.

Proof. Let \(A\) and \(D\) be disjoint closed subsets of \(X\). If either is empty, the open sets \(\varnothing\) and \(X\) separate them. Assume both are nonempty. A second-countable space, and each of its subspaces, is Lindelöf: every open cover has a countable subcover. Indeed, the subspace has a countable base. For each basis element that is contained in a member of the cover, choose one such cover member. The chosen members form a countable subcover, since each point lies in a basis element contained in a cover member.

For each \(a\in A\), regularity gives an open set \(U_a\) containing \(a\) whose closure misses \(D\). To see the closure condition, take disjoint open neighborhoods of \(a\) and \(D\); the closure of the first misses the second. The sets \(U_a\) cover \(A\), so the Lindelöf property of the subspace \(A\) gives a countable subcover \((U_i)\). Similarly, choose a countable open cover \((V_j)\) of \(D\) such that each \(\overline{V_j}\) misses \(A\). Define \[ U=\bigcup_i\left(U_i\setminus\bigcup_{j\leq i}\overline{V_j}\right), \qquad V=\bigcup_j\left(V_j\setminus\bigcup_{i\leq j}\overline{U_i}\right). \] Each term in these unions is open, because it removes only a finite union of closed sets from an open set. Each \(a\in A\) lies in some \(U_i\) and in none of the \(\overline{V_j}\), so \(a\in U\); likewise \(D\subseteq V\). If a point belonged to the \(i\)-th term of \(U\) and the \(j\)-th term of \(V\), then \(i\leq j\) would make the latter term exclude \(\overline{U_i}\), or \(j\leq i\) would make the former exclude \(\overline{V_j}\). Both are impossible. Thus \(U\) and \(V\) are disjoint open neighborhoods of \(A\) and \(D\), proving normality. \(\square\)

We now use Urysohn's lemma, the standard result that in a normal space, for disjoint closed sets \(F,G\), there is a continuous function \(f:X\to[0,1]\) with \(f=1\) on \(F\) and \(f=0\) on \(G\). The interval \([0,1]\) has its usual subspace topology from \(\mathbb{R}\). The lemma is used here in this precise form.

The Countable Family from Nested Base Pairs

Theorem: If \(X\) is regular and second-countable, there is a countable family of continuous functions \(f:X\to[0,1]\) such that (1) any two distinct points have different values under at least one function, and (2) for every open \(O\subseteq X\), $$ O=\bigcup_{f\in\mathcal{F}}\{x\in X:f(x)>0\}. $$ Here \(\mathcal{F}\) is the subfamily of functions selected for base pairs whose outer member is contained in \(O\).

Proof. Fix a countable base \(\mathcal{B}\), and consider the set of ordered pairs \[ \mathcal{P}=\{(B,C)\in\mathcal{B}\times\mathcal{B}:\overline{B}\subseteq C\}. \] It is countable because it is a subset of the countable set \(\mathcal{B}\times\mathcal{B}\). For every \((B,C)\in\mathcal{P}\), the sets \(\overline{B}\) and \(X\setminus C\) are disjoint and closed. Normality and Urysohn's lemma give a continuous \(f_{B,C}:X\to[0,1]\) such that \(f_{B,C}=1\) on \(\overline{B}\) and \(f_{B,C}=0\) on \(X\setminus C\). Thus \(\{x:f_{B,C}(x)>0\}\subseteq C\).

We verify that there are enough such pairs. Given \(x\in O\) with \(O\) open, choose \(C\in\mathcal{B}\) with \(x\in C\subseteq O\). Regularity gives an open \(W\) containing \(x\) with \(\overline{W}\subseteq C\): separate \(x\) from the closed set \(X\setminus C\), and take the closure of the neighborhood of \(x\). Choose \(B\in\mathcal{B}\) with \(x\in B\subseteq W\). Then \(\overline{B}\subseteq\overline{W}\subseteq C\), so \((B,C)\in\mathcal{P}\), and \(f_{B,C}(x)=1\). Because \(C\subseteq O\), this function is zero outside \(O\). We have shown that every \(x\in O\) belongs to the positive set of a function from a pair with \(C\subseteq O\), while every such positive set is contained in \(O\). This proves the stated union formula.

Finally, if \(x\ne y\), the \(T_1\) property makes \(O=X\setminus\{y\}\) open and it contains \(x\). The construction gives a function with \(f(x)=1\) and \(f(y)=0\). It therefore separates \(x\) and \(y\). If \(X\) is empty, the empty family satisfies both conclusions, so that case is included as well. \(\square\)

The union formula also says that the family detects the topology. Each \(f\) is continuous, so its inverse images of open subsets of \([0,1]\) are open in \(X\). Conversely, \((0,1]\) is open in \([0,1]\), and the formula writes every open subset of \(X\) as a union of sets \(f^{-1}((0,1])\). Thus the original open sets can be recovered from the functions.

Worked Constructions

Worked Example: Nested Rational Intervals in the Real Line

In \(\mathbb{R}\) with its standard topology, rational-endpoint open intervals form a countable base. Consider \(B=(0,2)\) and \(C=(-1,3)\). Their closures satisfy \(\overline{B}=[0,2]\subset C\). A function for this pair is $$ f(x)= \begin{cases} 0,&x\leq -1,\\ x+1,&-1\leq x\leq 0,\\ 1,&0\leq x\leq 2,\\ 3-x,&2\leq x\leq 3,\\ 0,&x\geq 3. \end{cases} $$

The formulas agree at every shared endpoint: the values at \(-1,0,2,3\) are respectively \(0,1,1,0\) from either adjoining piece. Each piece is continuous, so the matching endpoint values make \(f\) continuous on \(\mathbb{R}\). For \(x=1\), the middle piece gives \(f(1)=1\); for \(x=4\), the last piece gives \(f(4)=0\). Thus these two points are separated. Also, \(f(x)=0\) whenever \(x\notin C\), as required for the pair.

For a function positive near \(1\) but zero outside \(O=(0,2)\), take \(B=(1/2,3/2)\) and \(C=(1/4,7/4)\). Then \(\overline{B}=[1/2,3/2]\subset C\subset O\). The function from this pair equals \(1\) at \(1\), and equals \(0\) at every point outside \(C\), hence outside \(O\). This illustrates how the pair chosen for open-set detection can be made to lie inside the specified open set.

Worked Example: A Finite Discrete Space

Let \(X=\{a,b,c\}\) have the discrete topology, so every subset is open. Use the base of singletons \(\mathcal{B}=\{\{a\},\{b\},\{c\}\}\). Each singleton is closed as well as open, so \((\{a\},\{a\})\), \((\{b\},\{b\})\), and \((\{c\},\{c\})\) are suitable pairs. Their functions take value \(1\) on the named singleton and \(0\) elsewhere. In the order \(a,b,c\), their values are $$ \begin{array}{c|ccc} & a&b&c\\ \hline f_{\{a\},\{a\}}&1&0&0\\ f_{\{b\},\{b\}}&0&1&0\\ f_{\{c\},\{c\}}&0&0&1 \end{array} $$

For example, the first and second functions distinguish \(a\) from \(b\), since their values at those points are \(1,0\) and \(0,1\), respectively. For \(O=\{a,c\}\), the functions associated with \(\{a\}\) and \(\{c\}\) have positive sets \(\{a\}\) and \(\{c\}\); their union is exactly \(O\). For \(O=\varnothing\), the union of no positive sets is \(\varnothing\). Thus the formula works for these sets, including the empty open set.

Worked Example: A Convergent Sequence with Its Limit

Let \(X=\{0\}\cup\{1/n:n\in\mathbb{N}\}\) with the subspace topology inherited from \(\mathbb{R}\). Consider the relative open sets \(B=X\cap(-1/8,1/4)\) and \(C=X\cap(-1/4,1/2)\). We have \(B=\{0\}\cup\{1/n:n\geq5\}\) and \(C=\{0\}\cup\{1/n:n\geq3\}\). The closure of \(B\) in \(X\) is \(B\): its complement in \(X\) is the finite set \(\{1,1/2,1/3,1/4\}\), whose points are isolated and therefore form an open set. Hence \(\overline{B}=B\subset C\).

Define a continuous function on \(\mathbb{R}\) by $$ F(t)= \begin{cases} 0,&t\leq -1/4,\\ 8(t+1/4),&-1/4\leq t\leq -1/8,\\ 1,&-1/8\leq t\leq 1/4,\\ 2-4t,&1/4\leq t\leq 1/2,\\ 0,&t\geq 1/2. \end{cases} $$ The formulas agree at \(-1/4,-1/8,1/4,1/2\), with values \(0,1,1,0\), respectively. Restrict \(F\) to \(X\) to obtain a continuous function \(f:X\to[0,1]\). We have \(f(0)=1\) and \(f(1)=0\). Also, \(f(1/2)=2-4(1/2)=0\), \(f(1/3)=2-4/3=2/3\), and \(f(1/4)=1\). For \(n\geq5\), \(1/n<1/4\), so \(f(1/n)=1\). The values confirm that \(f\) is positive on points of \(C\) and zero outside \(C\), including at \(1\) and \(1/2\). In particular, this function detects the neighborhood \(X\setminus\{1\}\) of \(0\): its positive set contains \(0\) and is contained in that neighborhood.

Check: the decreasing-piece calculation at \(1/3\) gives \(2-4/3=2/3>0\), while at \(1/2\) it gives \(2-2=0\). These values match the claimed inclusion of the positive set in \(C\).

A Common Pitfall

Common mistake: choosing any two base elements that contain the desired point and assuming Urysohn's lemma applies. For instance, in \(\mathbb{R}\), take \(B=(0,2)\) and \(C=(1,3)\). Although both contain \(1.5\), the required closed sets intersect: $$ \overline{B}\cap(\mathbb{R}\setminus C)=[0,2]\cap\bigl((-\infty,1]\cup[3,\infty)\bigr)=[0,1]\ne\varnothing. $$ They are not disjoint, so Urysohn's lemma does not give a function equal to \(1\) on \(\overline{B}\) and \(0\) outside \(C\). The correction is to choose a genuinely nested pair, such as \(B=(0,2)\) and \(C=(-1,3)\), for which \([0,2]\subset(-1,3)\). The closure containment is the condition that makes the construction work.

Check Your Understanding

Use the nested-pair construction and verify the closure containment whenever you select a pair.

  1. In \(\mathbb{R}\), choose rational-endpoint intervals \(B,C\) with \(1\in B\), \(\overline{B}\subset C\), and \(C\subset(0,2)\). What does the associated function do at \(1\) and outside \((0,2)\)?
  2. Explain why the family of ordered pairs from a countable base is countable.
  3. In the theorem, why is \(X\setminus\{y\}\) open when \(X\) is \(T_1\)? How does this give a function separating \(x\ne y\)?
  4. For \(X=\{0\}\cup\{1/n:n\in\mathbb{N}\}\), find a relative open neighborhood of \(0\) that excludes \(1\), and describe what the detecting function must satisfy.
  5. Explain why the set \((0,1]\) is open in the subspace \([0,1]\), and how that fact recovers open sets from the functions.