ProofComplete

Proof of Urysohn Metrization Theorem

We give a complete proof that every second-countable regular Hausdorff space is metrizable by constructing an explicit embedding into the Hilbert cube. This proof synthesizes several fundamental results: the Lindel"of property, normality, the Urysohn lemma, and the theory of product topologies.


Statement

Theorem7.13Urysohn Metrization Theorem

Let XX be a T3T_3 second-countable space. Then XX is metrizable. Specifically, XX embeds as a subspace of the metrizable space [0,1]ω[0, 1]^{\omega} (the Hilbert cube with the product metric).


Complete Proof

Proof

Let B={B1,B2,}\mathcal{B} = \{B_1, B_2, \ldots\} be a countable basis for XX.

Part A: XX is normal.

Step A1: XX is Lindel"of. Let {Uα}\{U_\alpha\} be an open cover. For each α\alpha and each xUαx \in U_\alpha, choose Bn(x,α)BB_{n(x,\alpha)} \in \mathcal{B} with xBn(x,α)Uαx \in B_{n(x,\alpha)} \subseteq U_\alpha. The collection {Bn(x,α)}\{B_{n(x,\alpha)}\} is a (possibly redundant) subcover indexed by a subset of N\mathbb{N}, hence countable. For each such BkB_k in this subcover, choose one αk\alpha_k with BkUαkB_k \subseteq U_{\alpha_k}. Then {Uαk}\{U_{\alpha_k}\} is a countable subcover.

Step A2: XX is normal. This follows from Theorem 7.5: a regular Lindel"of space is normal.

Part B: Constructing separating functions.

Step B1: Define the set of "good pairs": P={(m,n)N2:BmBn}.\mathcal{P} = \{(m, n) \in \mathbb{N}^2 : \overline{B_m} \subseteq B_n\}.

This is a countable set. Since XX is regular: for any xx and open UxU \ni x, there exists BnBB_n \in \mathcal{B} with xBnUx \in B_n \subseteq U, and by regularity there exists an open VV with xVVBnx \in V \subseteq \overline{V} \subseteq B_n, and a BmB_m with xBmVx \in B_m \subseteq V. So BmVBn\overline{B_m} \subseteq \overline{V} \subseteq B_n, and (m,n)P(m, n) \in \mathcal{P}.

Step B2: For each (m,n)P(m, n) \in \mathcal{P}, the sets Bm\overline{B_m} and XBnX \setminus B_n are disjoint closed sets (using BmBn\overline{B_m} \subseteq B_n). By the Urysohn lemma (applicable since XX is normal), there exists a continuous function: gm,n:X[0,1],gm,nBm=0,gm,nXBn=1.g_{m,n}: X \to [0, 1], \quad g_{m,n}|_{\overline{B_m}} = 0, \quad g_{m,n}|_{X \setminus B_n} = 1.

Enumerate P\mathcal{P} as (m1,n1),(m2,n2),(m_1, n_1), (m_2, n_2), \ldots and write fk=gmk,nkf_k = g_{m_k, n_k}.

Part C: The embedding.

Step C1: Define F:X[0,1]ωF: X \to [0, 1]^{\omega} by: F(x)=(f1(x),f2(x),f3(x),).F(x) = (f_1(x), f_2(x), f_3(x), \ldots).

On [0,1]ω[0, 1]^{\omega}, use the metric: d(a,b)=k=1akbk2k.d(\mathbf{a}, \mathbf{b}) = \sum_{k=1}^{\infty} \frac{|a_k - b_k|}{2^k}.

This metric induces the product topology on [0,1]ω[0, 1]^{\omega}.

Step C2: FF is injective. Let xyx \neq y. Since XX is T1T_1, {y}\{y\} is closed. Set U=X{y}U = X \setminus \{y\}. By Step B1, there exists (m,n)P(m, n) \in \mathcal{P} with xBmBmBnUx \in B_m \subseteq \overline{B_m} \subseteq B_n \subseteq U. Then yBny \notin B_n, so gm,n(x)=0g_{m,n}(x) = 0 and gm,n(y)=1g_{m,n}(y) = 1. Hence F(x)F(y)F(x) \neq F(y).

Step C3: FF is continuous. Each coordinate function πkF=fk\pi_k \circ F = f_k is continuous. By the universal property of the product topology, FF is continuous.

Step C4: FF is an open map onto F(X)F(X). Let UXU \subseteq X be open and xUx \in U. By Step B1, there exists (mk,nk)P(m_k, n_k) \in \mathcal{P} with xBmkBnkUx \in B_{m_k} \subseteq B_{n_k} \subseteq U. Then fk(x)=0f_k(x) = 0 and fk(y)=1f_k(y) = 1 for all yXUy \in X \setminus U.

Consider the set W={a[0,1]ω:ak<1}W = \{\mathbf{a} \in [0,1]^{\omega} : a_k < 1\}. This is open in [0,1]ω[0,1]^{\omega}.

We have F(x)WF(x) \in W (since fk(x)=0f_k(x) = 0). For any zF1(W)Xz \in F^{-1}(W) \cap X: fk(z)<1f_k(z) < 1, which means zXBnkz \notin X \setminus B_{n_k}, i.e., zBnkUz \in B_{n_k} \subseteq U.

Thus F1(W)UF^{-1}(W) \subseteq U, so F(U)F(X)WF(U) \supseteq F(X) \cap W, an open set in F(X)F(X) containing F(x)F(x).

This shows F(U)F(U) is open in F(X)F(X).

Conclusion. F:XF(X)F: X \to F(X) is a continuous bijection that is also an open map, hence a homeomorphism. Since F(X)[0,1]ωF(X) \subseteq [0, 1]^{\omega} and [0,1]ω[0,1]^{\omega} is metrizable (with metric dd), F(X)F(X) is metrizable, and therefore XX is metrizable.


Key Observations

RemarkRole of Each Hypothesis
  • T1T_1 (from T3T_3): Ensures singletons are closed, needed for injectivity of FF.
  • Regularity (from T3T_3): Ensures the existence of good pairs (Bm,Bn)(B_m, B_n) with BmBn\overline{B_m} \subseteq B_n.
  • Second countability: Ensures P\mathcal{P} is countable, so the embedding is into a countable product [0,1]ω[0,1]^{\omega}.
  • Normality (derived from regular + Lindel"of): Enables the Urysohn lemma.
ExampleMetrizing $S^n$

The sphere SnRn+1S^n \subseteq \mathbb{R}^{n+1} is second-countable (subspace of Rn+1\mathbb{R}^{n+1}) and regular (Hausdorff and locally compact). By the Urysohn metrization theorem, SnS^n is metrizable. Of course, SnS^n already carries the metric inherited from Rn+1\mathbb{R}^{n+1}, but the theorem confirms that any second-countable regular topology on SnS^n is metrizable -- the result is topology-intrinsic, not dependent on the ambient Euclidean space.

RemarkThe Hilbert Cube as a Universal Space

The Hilbert cube [0,1]ω[0, 1]^{\omega} is a universal separable metrizable space: every separable metrizable space embeds in it. Since second-countable metrizable spaces are separable, the Urysohn embedding X[0,1]ωX \hookrightarrow [0,1]^{\omega} is optimal. The Hilbert cube is compact, metrizable, and infinite-dimensional, serving as the "home" for all separable metrizable spaces.