This section uses the preceding Urysohn lemma and the usual classical set-theoretic background, including countable choices.
Motivation
A topology tells us which neighborhoods are available, without assigning distances. To recover a metric, we need enough continuous numerical measurements to distinguish nearby points from points outside a prescribed neighborhood. Urysohn functions provide individual measurements. Second countability lets us organize all the necessary measurements into one sequence, whose weighted sum is finite.
There are two tasks after writing that sum. Distinct points must have positive distance, and its balls must reproduce the original topology. Point separation proves only the first. The stronger requirement that some measurement separates a point from the complement of any neighborhood will prove the second. Keeping these tasks distinct prevents a common incomplete metrization argument.
The theorem is a sufficient condition for metrizability. An uncountable discrete space already has a metric and no countable base. We will also see that a countable dense subset, even together with countable neighborhood bases at individual points, cannot replace second countability.
Definitions
Definition 13.1.1
(Bases and countability axioms)
A base is a family of open sets such that every open neighborhood of every point contains a member of the family containing that point. A space is second countable if it has a countable base. It is first countable if every point has a countable family of open neighborhoods refining all its neighborhoods. A countable base immediately gives such a local family, by retaining the members containing the point.
Definition 13.1.2
(Density, separability and the Lindelöf property)
A subset is dense if its closure is the whole space; equivalently, it meets every nonempty open set. A space is separable if it has a countable dense subset, and Lindelöf if every open cover has a countable subcover. Countable here includes finite. Empty spaces can be treated directly, so the proofs may assume nonempty spaces when choosing points.
Definition 13.1.3
(Compatible metrics and regularity)
A metric is compatible with a topology if its open balls generate exactly that topology. Recall our convention: regular means together with separation of a point and a disjoint closed set by disjoint open neighborhoods. Equivalently, is and whenever with open, there is open with . To prove this equivalence, separate from ; conversely use and .
Theorem/proposition
Theorem 13.1.4
(From a countable base to normality)
Every second-countable space is Lindelöf. Every regular Lindelöf space is normal, under the convention that normality includes . Consequently, every regular second-countable space is normal.
Theorem 13.1.5
(Urysohn metrization theorem)
Every regular second-countable space admits a compatible metric. More precisely, there is a countable family of continuous functions such that
is a compatible metric. For every with open, some equals one at and vanishes outside .
Theorem 13.1.6
(Countability inside metric spaces)
A metrizable space is second countable if and only if it is separable. This equivalence is specific to metrizable spaces; in a general space second countability still implies separability, but the converse can fail.
Proof sketch or proof idea
Proof
(Countable bases give normality)
Let be a countable base and an open cover. For each basis member contained in some member of , select one such covering set. These selected sets cover : given , first choose a covering neighborhood and then a basis member inside it. There are at most countably many selections. This proves Lindelöfness without supposing that the original cover was countable.
A closed subspace of a Lindelöf space is Lindelöf. Write a relative open cover as , append , and apply the ambient property. Removing the appended set leaves a countable cover of .
Now let be disjoint closed sets in a regular Lindelöf space. Regularity gives neighborhoods of each point of whose closures miss , and neighborhoods of each point of whose closures miss . The closed-subspace observation extracts countable families respectively covering , with
Pad finite lists by empty sets. If either closed set is empty, separation is immediate. Otherwise set
Each term is open, because only finitely many closed sets are removed. No removed closure meets the closed set that the corresponding family covers, so and . If a point belonged to the th term of and the th term of , then makes the first term exclude , while makes the second exclude . At least one inequality holds. Thus are disjoint, proving normality.
Proof
(A weighted sum of Urysohn functions)
For metrization, enumerate a countable base as . For every pair with , Urysohn's lemma gives a continuous function equal to one on and zero on . These are disjoint closed sets, precisely the hypotheses needed for that lemma. The pairs form a countable set. List their functions as , padding a finite or empty list with zero functions.
Why are there enough such pairs? Given , first take . Regularity shrinks a neighborhood of so that . A basis member with then satisfies . The corresponding function is one at and zero outside . In particular, for , apply this with , which is open by . Some function distinguishes from .
The displayed series for converges, with its tail after term bounded by . Symmetry and nonnegativity hold termwise. If , all coordinate differences vanish, so separation gives . Each summand obeys the triangle inequality; sum the finite inequalities and take limits to obtain it for .
We must now compare topologies in both directions. Fix . Each finite partial sum is a continuous real function of , and the tail bound is uniform in . Hence is continuous in the original topology. Explicitly, choose a tail smaller than one third of a desired error, then use continuity of the finite sum; this proves continuity without assuming that the domain already has a metric. Consequently every -ball is originally open.
Conversely, if , choose vanishing outside . Then
Every original neighborhood therefore contains a -ball about each of its points, so is -open. Both topologies agree.
Proof X-ray
The measurement that detects a neighborhood
The family is built from nested basis pairs, rather than merely from pairs of distinct points. Point separation makes the sum a metric. The extra vanishing condition outside any prescribed open set makes that metric compatible. The geometric weights address convergence; they do not supply either separation property by themselves.
Proof
(Separability and countable bases in metric spaces)
Finally, choose one point from each nonempty member of a countable base. The resulting countable set meets every nonempty open set, proving separability. In the other direction let be countable dense in a metric space. Balls with centers in and positive rational radii form a countable base. Indeed, for , choose with , choose with , and choose rational with
Then . The interval for is nonempty, and the triangle inequality supplies the inclusion.
Worked examples
Worked example 13.1.7
(The countable cube and moving coordinates)
On , the metric generates the product topology. Finite partial sums are continuous for the product topology and the tails are uniformly small, so its balls are product-open. Conversely, a basic product neighborhood restricts only finitely many coordinates; sufficiently small distance forces each of those coordinates into its prescribed interval, using .
Let have its only nonzero coordinate, equal to one, in position . Then . The supremum distance stays one. Product topology permits a discrepancy to move out through later coordinates; uniform control of every coordinate is a different topology.
Concept lensGeometric
Metrization as an embedding into the countable cube
The functions of the theorem combine into , and the metric is exactly the cube metric of this example pulled back along . The metrization proof therefore shows that is a homeomorphism onto its image. Conversely, every subspace of the cube is regular and second countable. So the regular second-countable spaces are, up to homeomorphism, exactly the subspaces of .
Worked example 13.1.8
(A large discrete metric space)
On an uncountable set put for and zero otherwise. Balls of radius are singletons, so this is the discrete topology. Any base must contain every singleton: it must refine the singleton neighborhood of its sole point. Thus no countable base exists. This space is metrizable, regular and normal; the theorem's hypotheses do not describe every metrizable space.
Worked example 13.1.9
(The Sorgenfrey line: many local bases, no global countable base)
Give the topology with basis , . These intervals are clopen: their complements are unions of basis intervals. Singletons are closed, and the clopen base gives the shrinking property, so the space is regular. At each , the sets are a countable neighborhood base. The rationals are dense, because every nonempty basis interval contains a rational.
Suppose were a countable base. For each real , choose with . This set has minimum . Different therefore require different members of , an impossibility. The space is separable but not second countable; the metric-space equivalence just proved shows it is not metrizable.
Worked example 13.1.10
(A compact Hausdorff corollary)
A second-countable compact Hausdorff space is metrizable. Compact Hausdorff spaces are normal by the earlier separation proof, hence regular in our convention; apply the theorem. Compactness alone does not provide a countable base, so that hypothesis must still be checked. For a subspace of Euclidean space, rational-coordinate balls give a countable relative base, making this corollary applicable to compact surfaces already embedded there.
Common mistakes
Common mistake
(Separable is not second countable)
Separability chooses points meeting open sets; second countability chooses neighborhoods refining open sets. The Sorgenfrey example shows why these are different demands. First countability only organizes the neighborhoods of one point at a time, without limiting how many different families are needed.
Common mistake
(An injective map into the cube need not be an embedding)
An infinite sum of continuous functions need not be continuous. Here a summable uniform bound supplies the missing justification. Likewise, injectivity of a continuous map into a cube does not alone make it an embedding: the inverse onto its image must be continuous. The neighborhood-detecting functions prove precisely that missing part.
Common mistake
(Regularity alone does not separate closed sets)
Normality was obtained before using Urysohn's lemma. Regularity by itself does not authorize closed-set separation. Finally, the constructed metric is generally not canonical and carries no promised completeness, geodesic or length interpretation.
Counterexample mode
Hausdorff cannot replace regular
Claim: every Hausdorff second-countable space is metrizable. Let and give the topology with basis all open intervals together with all sets . Intervals and sets with rational already form a base, so the space is second countable, and it is Hausdorff because it refines the usual topology. The set is closed, since is the union of the usual open set and the basic set . Now let and be open. Then contains some . Choose with ; since , the set contains an interval with , and this interval contains points outside lying in , hence in . So and cannot be separated: the space is not regular and therefore not metrizable. Regularity, not the Hausdorff property, is the hypothesis the proof uses when it shrinks to .
Summary
A countable base gives Lindelöfness; regularity and Lindelöfness give normality; nested basis pairs and Urysohn give countably many neighborhood-detecting functions. Their uniformly weighted distance sum recovers the topology. Within metric spaces, separability and second countability agree, while the lower-limit line separates these notions outside that setting.
Exercises
Quick check
(Prove that every subspace of a second-countable space is second countable, and that the product of two second-countable spaces is second countable. Deduce metrizability for a regular subspace once the inherited regularity has been checked.)
Solution · Answer
If is a base of , the nonempty sets form a countable base of : intersect a refining ambient neighborhood with . Products form a countable base of the product by its definition. Regularity is hereditary: for with ambient-open, choose ambient-open with ; then . The property is hereditary too. Thus every subspace of a regular second-countable space satisfies the metrization hypotheses.
Quick check
(Replace in the countable-cube metric by positive numbers with . Prove compatibility. What fails if ?)
Solution · Answer
The tail tends to zero uniformly in the two cube points. Finite sums are continuous, so metric balls are product-open. A restriction on finitely many indices is forced by . Positivity distinguishes different points. If , the distance between the all-zero and all-one points diverges, so the formula does not even define a finite metric on the entire cube.
Quick check
(Show that cube points with finitely many nonzero rational coordinates form a countable dense subset. Explain why zero is allowed in this description.)
Solution · Answer
For each finite support there are countably many rational assignments, and there are countably many finite subsets of , so the family is countable. A nonempty basic product-open set restricts finitely many coordinates to nonempty relative open subsets of . Choose a rational in each such subset and put zero in every other coordinate. This gives a point of the family in that neighborhood. An interval touching an endpoint still contains a permitted rational, including the endpoint when appropriate; zero outside the restricted coordinates is harmless.
Quick check
(Prove that a compact metrizable space is second countable, without assuming separability in advance. Use finite covers by small balls.)
Solution · Answer
For each positive integer , compactness extracts a finite -ball cover. Choose their centers and let be the union of these finite sets. It is countable. Every nonempty open set contains some ball ; for , one of the selected centers lies within of , hence inside the open set. Thus is dense. Apply the metric separability equivalence. The empty space is immediate.
Quick check
(Let two metrics be formed from the same separating functions , with two different positive summable weight sequences. Prove that the identity between the two metrics is uniformly continuous in both directions, even when is not compact.)
Solution · Answer
Write and similarly . Given , choose with . Put . Each difference is at most , so the head of is at most . If , the head and tail together are less than . This bound is independent of , proving uniform continuity. Interchange for the reverse direction. The argument gives no comparison with a metric built from a completely different family without further information.