Separation Axioms — T1, Hausdorff, Normal

⚡ Key Takeaways

Separation axioms classify topological spaces by how well distinct points and closed sets can be "separated" by open sets. T₁ ⟺ points are closed; T₂ (Hausdorff) ⟺ distinct points have disjoint neighborhoods (guaranteeing unique limits); T₄ (Normal + T₁) ⟺ disjoint closed sets have disjoint open neighborhoods (enabling Urysohn's lemma and Tietze extension). The strict hierarchy T₄ ⟹ T₃ ⟹ T₂ ⟹ T₁ ⟹ T₀ is non-reversible. Core toolkit of point-set topology, developed 1907–1930s by Fréchet, Hausdorff, Tietze, and Urysohn.

💡 The Core Insight

"Separation" is the dial that controls how much continuous-function structure a space supports. Each rung up the ladder buys you something concrete: T₁ makes singletons closed; T₂ makes limits unique and diagonals closed; T₃ separates points from closed sets; T₄ separates closed sets from each other — which is exactly the hypothesis Urysohn's Lemma needs to manufacture continuous functions out of pure topology. Continuous real functions, not open sets, are the real currency of separation.

DefinitionT₀ and T₁ (Kolmogorov and Fréchet)

Let $(X,\tau)$ be a topological space.

T₀ (Kolmogorov): for every pair of distinct points $x\neq y$ , at least one has an open neighborhood excluding the other.
T₁ (Fréchet): for every pair $x\neq y$ there exist open sets $U,V$ with $x\in U$ , $y\notin U$ and $y\in V$ , $x\notin V$ .

The defining equivalence for T₁:

1

\quad X \text{ is } T_1 \iff \forall x\in X,\ \{x\} \text{ is closed} \iff \{x\} = \bigcap_{x\in U,\, U\in\tau} U .

Derivation hint: if every singleton is closed, $X\setminus\{y\}$ is an open neighborhood of any $x\neq y$ avoiding $y$ ; conversely intersect all open sets containing $x$ to recover $\{x\}$ as a closed set.

DefinitionT₂ — Hausdorff

$(X,\tau)$ is Hausdorff (T₂) if for all $x\neq y$ there exist disjoint open sets $U\ni x$ , $V\ni y$ :

2

\quad \forall\, x\neq y\ \exists\, U,V\in\tau:\quad x\in U,\ y\in V,\ U\cap V=\varnothing .

Derivation hint: this is "separated by neighborhoods"; strengthen the T₁ condition by demanding the two neighborhoods be disjoint, not merely point-excluding. An equivalent, exceptionally useful characterization via the diagonal $\Delta = \{(x,x):x\in X\}\subseteq X\times X$ :

3

\quad X \text{ is Hausdorff} \iff \Delta \text{ is closed in } X\times X .

Derivation hint: $(x,y)\notin\Delta$ means $x\neq y$ ; a basic open box $U\times V$ around $(x,y)$ misses $\Delta$ iff $U\cap V=\varnothing$ .

DefinitionT₃ — Regular + T₁, and T₄ — Normal + T₁

A space is regular if every point can be separated from every closed set not containing it:

4

\quad \forall\, x\notin F\ (F\text{ closed})\ \exists\, U,V\in\tau:\quad x\in U,\ F\subseteq V,\ U\cap V=\varnothing .

Derivation hint: replace one of the two "points" in the Hausdorff condition by an arbitrary closed set. A space is normal if every two disjoint closed sets are separated by disjoint open sets:

5

\quad \forall\, F_1,F_2 \text{ closed},\ F_1\cap F_2=\varnothing\ \exists\, U_1,U_2\in\tau:\quad F_i\subseteq U_i,\ U_1\cap U_2=\varnothing .

Derivation hint: replace both "points" of the Hausdorff condition by disjoint closed sets. Conventions: $T_3 = \text{regular} + T_1$ ; $T_4 = \text{normal} + T_1$ . (Some authors bake T₁ into "regular/normal"; one must always check the convention.) The hierarchy:

6

\quad T_4 \implies T_3 \implies T_2 \implies T_1 \implies T_0 .

Derivation hint: each implication uses the lower axiom to convert points into closed sets — T₁ makes $\{x\}$ closed, so normality applied to $\{x\},\{y\}$ gives Hausdorff; regularity applied to $\{x\},F$ gives Hausdorff via $F=\{y\}$ .

TheoremUrysohn's Lemma

A T₁ space $X$ is normal if and only if for every pair of disjoint closed sets $A,B$ there exists a continuous function

7

\quad f:X\to[0,1],\qquad f|_A\equiv 0,\quad f|_B\equiv 1 .

Derivation hint: one direction is the deep content (constructing $f$ ); the converse is easy — preimages $f^{-1}[0,\tfrac12)$ and $f^{-1}(\tfrac12,1]$ separate $A,B$ .

TheoremTietze Extension Theorem

$X$ is normal iff every continuous $f:A\to\mathbb{R}$ on a closed subset $A\subseteq X$ extends to a continuous $F:X\to\mathbb{R}$ with $F|_A=f$ ; moreover the bound is preserved:

8

\quad \sup_{x\in X}|F(x)| = \sup_{a\in A}|f(a)| .

Derivation hint: build $F$ as a uniformly convergent series of Urysohn functions correcting the error on $A$ in geometric $\tfrac{2}{3}$ -sized steps.

ProofProof of Urysohn's Lemma (construction sketch)

Let $A,B$ be disjoint closed sets in a normal space $X$ .

Step 1 — Dyadic shrinking. By normality choose open $U_{1/2}$ with $A\subseteq U_{1/2}$ and $\overline{U_{1/2}}\cap B=\varnothing$ . The key reformulation of normality used here:

9

\quad A\subseteq U \text{ open} \implies \exists V \text{ open}:\ A\subseteq V\subseteq \overline V\subseteq U .

Derivation hint: apply normality to disjoint closed sets $A$ and $X\setminus U$ .

Step 2 — Indexing by dyadic rationals. Recursively assign to each dyadic rational $q\in[0,1]\cap\mathbb{Z}[\tfrac12]$ an open set $U_q$ so that

10

\quad p < q \implies \overline{U_p}\subseteq U_q,\qquad A\subseteq U_0,\quad X\setminus B = U_1 .

Derivation hint: at each dyadic refinement insert a set between $U_p$ and $U_q$ using [9].

Step 3 — Define the function.

11

\quad f(x) = \inf\{\, q : x\in U_q\,\},\qquad f(x)=1 \text{ if } x\notin U_q\ \forall q<1 .

Derivation hint: $f$ rises from $0$ on $A$ to $1$ on $B$ as $x$ leaves successively larger $U_q$ .

Step 4 — Continuity. Continuity follows because the sublevel and superlevel sets are open:

12

\quad f^{-1}\big([0,a)\big) = \bigcup_{q<a} U_q,\qquad f^{-1}\big((a,1]\big) = \bigcup_{q>a}\big(X\setminus \overline{U_q}\big).

Derivation hint: $f(x)<a$ iff $x\in U_q$ for some dyadic $q<a$ ; $f(x)>a$ iff $x\notin\overline{U_q}$ for some dyadic $q>a$ , using the nesting in [10]. Both families are unions of open sets, hence open; since intervals $[0,a)$ and $(a,1]$ generate the topology of $[0,1]$ , $f$ is continuous. $\blacksquare$

∎

mathematician · Pavel Urysohn | 1898–1924

Pavel Samuilovich Urysohn, working in Moscow alongside Pavel Alexandrov, proved in 1924–25 the two pillars that bear his name: Urysohn's Lemma (normality manufactures continuous functions) and the Urysohn Metrization Theorem (a second-countable regular space is metrizable). These results revealed that the separation axioms were not bookkeeping but the precise topological substrate of analysis. Tragically Urysohn drowned while swimming off the coast of Brittany in 1924 at age 26; Alexandrov edited and published much of his work posthumously, and the metrization program they began culminated decades later in the Nagata–Smirnov–Bing theorems.

🧠 Intuition

Think of separation as resolution power, like a microscope. T₀ says no two points are literally indistinguishable. T₁ says you can isolate any single point — its singleton is "in focus" as a closed set. T₂ (Hausdorff) says any two points sit in non-overlapping fields of view, so a convergent net cannot smear across two limit values. Regularity (T₃) widens the field: not just points, but a point and a whole closed "wall" can be put in disjoint focus. Normality (T₄) is the strongest practical level: two disjoint walls can be insulated. Urysohn's lemma then says this insulation is so robust you can paint a continuous temperature gradient running 0°C on one wall to 1°C on the other — pure topology gives you analysis.

Figure

The separation hierarchy and what each axiom separates

```svg

```

ExampleVerifying Hausdorff for a metric space, with explicit radii

Given: $X=\mathbb{R}^2$ with the Euclidean metric; points $x=(0,0)$ and $y=(3,4)$ , so $d(x,y)=\sqrt{3^2+4^2}=5$ .

Goal: produce disjoint open neighborhoods.

Steps: Choose radius $r=\tfrac{d(x,y)}{2}=\tfrac52=2.5$ . Take open balls $U=B(x,2.5)$ and $V=B(y,2.5)$ . Suppose $z\in U\cap V$ . Then by the triangle inequality

13

\quad d(x,y)\le d(x,z)+d(z,y) < 2.5 + 2.5 = 5 = d(x,y),

Derivation hint: strict inequalities because $z$ lies strictly inside both balls; summing contradicts $d(x,y)=5$ . a contradiction, so $U\cap V=\varnothing$ .

Boxed answer: $\boxed{U=B((0,0),\,2.5),\ V=B((3,4),\,2.5)\text{ are disjoint open sets separating }x,y}$ . Every metric space is Hausdorff (indeed normal).

ExampleA T₁ space that is NOT Hausdorff: the cofinite topology

Given: $X=\mathbb{R}$ with the cofinite topology: open sets are $\varnothing$ and all $U$ with finite complement $X\setminus U$ .

T₁ check: singleton $\{a\}$ has complement $\mathbb{R}\setminus\{a\}$ , which is open (its complement $\{a\}$ is finite), so $\{a\}$ is closed. By [1], $X$ is T₁. ✓

Hausdorff check: take $x=0$ , $y=1$ and any open $U\ni 0$ , $V\ni 1$ . Then $X\setminus U$ and $X\setminus V$ are finite, so

14

\quad X\setminus(U\cap V) = (X\setminus U)\cup(X\setminus V) \text{ is finite} \implies U\cap V \text{ is infinite} \neq\varnothing .

Derivation hint: in an infinite cofinite space, any two nonempty open sets have cofinite, hence overlapping, content. So no disjoint neighborhoods exist.

Boxed answer: $\boxed{(\mathbb{R},\text{cofinite}) \text{ is } T_1 \text{ but not } T_2}$ — proving the implication $T_2\Rightarrow T_1$ is strict.

⚠️ Common mistake

"Normal implies Hausdorff." False as stated. Normal alone (every two disjoint closed sets separated) does not imply T₂ — an indiscrete space with $\geq 2$ points is vacuously normal (the only disjoint closed sets are $\varnothing$ and $X$ , trivially separable) yet not even T₁. You need the T₁ qualifier: $T_4 = \text{normal}+T_1 \Rightarrow T_2$ . Always state whether T₁ is included. Likewise "regular" without T₁ does not imply Hausdorff.

⚠️ Common mistake

"Separation axioms are hereditary and productive." Only the lower ones are. T₀, T₁, T₂, T₃ are hereditary (pass to subspaces) and productive (pass to products). Normality fails both: subspaces of normal spaces need not be normal, and products of normal spaces need not be normal. Hence T₄ is the fragile rung; this fragility motivates the entire metrization theory.

CounterexampleThe Sorgenfrey plane — normality is not productive

The Sorgenfrey line $\mathbb{R}_\ell$ (basis $[a,b)$ ) is normal (in fact Lindelöf + regular ⟹ normal). But its square $\mathbb{R}_\ell\times\mathbb{R}_\ell$ , the Sorgenfrey plane, is not normal. The anti-diagonal

15

\quad D = \{(x,-x): x\in\mathbb{R}\}

Derivation hint: $D$ is closed and discrete in the Sorgenfrey plane; split it into the rational and irrational sub-anti-diagonals. is a closed discrete subspace; the disjoint closed sets $D\cap(\mathbb{Q}^2$ -shifted $)$ and $D\cap(\text{irrationals})$ cannot be separated by disjoint open sets, by a Baire-category/cardinality argument. This is the canonical demonstration that the product of two normal (indeed paracompact, Lindelöf) spaces can fail to be normal.

CounterexampleA Hausdorff space that is not regular

Let $X=\mathbb{R}$ with the topology generated by the usual open sets together with declaring $K=\{1/n : n\in\mathbb{N}\}$ "closed-extra" via the basis $U\setminus K$ for usual open $U\ni 0$ . This "K-topology" $\mathbb{R}_K$ is Hausdorff, but the point $0$ and the closed set $K$ cannot be separated by disjoint open sets: every basic neighborhood of $0$ is $U\setminus K$ , and any open set containing $K$ must come arbitrarily close to $0$ . Thus $\boxed{\mathbb{R}_K \text{ is } T_2 \text{ but not } T_3}$ , showing $T_3\Rightarrow T_2$ is strict.

connection

Separation axioms sit at the crossroads of point-set topology and analysis. They are the hypotheses of the great metrization theorems — Urysohn Metrization Theorem (regular + second countable ⟹ metrizable) and Nagata–Smirnov Metrization Theorem (regular + $\sigma$ -locally finite basis ⟹ metrizable). Normality is the precise condition for Tietze Extension Theorem and the existence of Partitions of Unity on paracompact Hausdorff spaces, which in turn underpin Riemannian metrics and integration on Smooth Manifolds. Hausdorffness is essential to Compactness theory (compact subsets of Hausdorff spaces are closed; compact Hausdorff ⟹ normal) and to the uniqueness of limits in Nets and Filters. The diagonal characterization [3] links to Product Topology and to the definition of separatedness in Scheme Theory (algebraic geometry's analog of Hausdorff).

🔬 Where this appears

⏳ History

- 1906–1907 — Maurice Fréchet introduces metric and abstract "L-spaces" in his thesis; the T₁ condition (closed points) later bears his name (Fréchet space, in the separation sense).

- 1914 — Felix Hausdorff publishes Grundzüge der Mengenlehre, defining topological spaces via neighborhood axioms and including the separation property that becomes the namesake Hausdorff axiom (T₂).

- 1921 — Heinrich Tietze systematically names and orders the Trennungsaxiome T₁–T₅, giving the field its "T" notation and the framework of a graded hierarchy.

- 1923 — Tietze proves his extension theorem for metric spaces (later seen to characterize normality).

- 1924 — Pavel Urysohn proves Urysohn's Lemma and the Urysohn Metrization Theorem; he drowns off Brittany the same year.

- 1925 — Alexandrov & Urysohn (posthumous) develop compactness and bicompactness theory, tying it to separation.

- 1930 — Andrey Tychonoff introduces complete regularity (T₃½, Tychonoff spaces) and proves products of compact spaces are compact, clarifying which separation axioms are productive.

- 1937 — Eduard Čech / Marshall Stone construct the Stone–Čech compactification, available exactly for Tychonoff spaces, embedding separation theory into compactification theory.

- 1948–1951 — Nagata, Smirnov, Bing independently complete the general metrization theorem, characterizing metrizable spaces via regularity plus a

\sigma

-locally-finite basis — the capstone of the separation-axiom program.

open-problem

The Normal Moore Space Conjecture asked whether every normal Moore space is metrizable. The astonishing resolution: the question is independent of ZFC. Under the assumption $V=L$ (or just $2^{\aleph_0} < 2^{\aleph_1}$ ) every normal Moore space is metrizable; but consistently (assuming a measurable cardinal / certain forcing axioms) there exist normal, non-metrizable Moore spaces. This places fine separation questions squarely in set-theoretic topology, where properties like normality of products (Dowker spaces, the Sorgenfrey plane, Michael line) depend on axioms beyond ZFC such as CH, MA, or the existence of large cardinals. Open and active themes include: characterizing exactly which products of normal spaces are normal, the structure of Dowker spaces (normal spaces whose product with $[0,1]$ is not normal), and the boundary between collectionwise normality and ordinary normality.

Separation Axioms — T1, Hausdorff, Normal

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also