A topological space records which subsets count as open, but many arguments in analysis need more than open sets. They need the ability to separate two incompatible pieces of information by disjoint neighbourhoods, and later by a [continuous function](/page/Continuous%20Function). Normality is the separation axiom that makes this possible for closed sets. The point is not cosmetic. If two closed sets cannot be pulled apart, then a function that should be zero on one set and one on the other may not exist. Partitions of unity, extension theorems, and many compactness arguments depend on the same local-to-global manoeuvre: isolate closed conditions without disturbing the rest of the space. [example: Separating Two Closed Intervals in $\mathbb{R}$] Let $X=\mathbb{R}$ with its usual topology, and let $A=[0,1]$ and $B=[3,4]$. The sets $A$ and $B$ are closed intervals, hence closed in the usual topology, and they are disjoint because every $a\in A$ satisfies $a\leq 1$ while every $b\in B$ satisfies $b\geq 3$. The [open set](/page/Open%20Set) $U=(-1,2)$ contains $A$, since $0>-1$ and $1<2$. The open set $V=(2,5)$ contains $B$, since $3>2$ and $4<5$. Also $U\cap V=\varnothing$: if $x\in U\cap V$, then $x<2$ from $x\in U$ and $x>2$ from $x\in V$, which is impossible. The set-to-set distance is exactly $2$. For any $a\in[0,1]$ and $b\in[3,4]$, we have $a\leq 1<3\leq b$, so $|a-b|=b-a$. Since $b\geq 3$ and $a\leq 1$, \begin{align*} |a-b|=b-a\geq 3-1=2. \end{align*} Therefore $2$ is a lower bound for $\{|a-b|:a\in A,\ b\in B\}$. Taking $a=1$ and $b=3$ gives \begin{align*} |1-3|=2, \end{align*} so the lower bound is attained, and hence \begin{align*} \operatorname{dist}(A,B)=\inf\{|a-b|:a\in A,\ b\in B\}=2. \end{align*} This positive gap is what the open buffers $(-1,2)$ and $(2,5)$ are exploiting: each closed interval is surrounded by an open neighbourhood, and the two neighbourhoods still do not meet. [/example] In metric spaces this behaviour is so familiar that it can become invisible. Normal spaces isolate exactly the topological content of that separation phenomenon, without mentioning distances. ## Definition A topological space already tells us what open and closed sets are. Normality asks for an additional promise: whenever two closed conditions are mutually exclusive, the topology has enough open sets to keep them apart. [definition: Normal Topological Space] A topological space $(X,\tau)$ is a normal topological space if for every pair of closed sets $A,B \subset X$ with $A \cap B = \varnothing$, there exist open sets $U,V \in \tau$ such that \begin{align*} A \subset U, \quad B \subset V, \quad U \cap V = \varnothing. \end{align*} [/definition] This definition is a child of the general notion of a [topological space](/page/Topological%20Space): the underlying object is still just $(X,\tau)$, but the topology is required to have a stronger separation property. The condition is about closed sets, not arbitrary subsets, because closed sets encode constraints that are stable under limits. ## Separation Axioms and Neighbourhoods Normality is easiest to use once we separate the page-topic definition from the auxiliary language around it. The first auxiliary notion names the open sets that surround a set, since normality is a statement about choosing two such surrounding sets with no overlap. [definition: Open Neighbourhood of a Set] Let $(X,\tau)$ be a topological space and let $A \subset X$. An open neighbourhood of $A$ is an open set $U \in \tau$ such that $A \subset U$. [/definition] With this terminology, normality says that disjoint closed sets have disjoint open neighbourhoods. The definition also explains why normality strengthens the Hausdorff intuition: points are small closed sets in many spaces, but closed sets may be much larger and harder to separate. To turn that intuition into a theorem, we must know when points themselves are closed enough for normality to apply. [definition: $T_1$ Topological Space] A topological space $(X,\tau)$ is a $T_1$ topological space if for every pair of distinct points $x,y \in X$, there exists an open set $U \in \tau$ such that $x \in U$ and $y \notin U$. [/definition] The next question is whether closed-set separation recovers the usual separation of distinct points. In a $T_1$ space, singleton sets are closed, so normality can be applied to $\{x\}$ and $\{y\}$. This gives the first structural consequence of the definition and places normality among the standard separation axioms. [quotetheorem:9166] This theorem places normality among the separation axioms. It is stronger than Hausdorffness under the usual $T_1$ convention, because it separates all disjoint closed sets rather than only two singleton sets. [remark: Convention on Normality] Some authors build the $T_1$ condition into the phrase "normal space." On this page, normality means only closed-set separation, and the $T_1$ hypothesis is stated separately whenever it is needed. [/remark] ## Metric Sources of Normality The most reliable supply of normal spaces comes from metric spaces. A metric gives a numerical way to measure how far a point is from a [closed set](/page/Closed%20Set), and these distance functions turn closed-set separation into inequalities. To use this mechanism, we need the distance from a point to a set. The definition is metric rather than purely topological, and it is exactly the device that later disappears from the final topological statement. [definition: Distance from a Point to a Set] Let $(X,d)$ be a [metric space](/page/Metric%20Space), and let $A \subset X$ be nonempty. The distance from a point to $A$ is the function $\operatorname{dist}(\cdot,A):X\to[0,\infty)$ defined by \begin{align*} \operatorname{dist}(x,A) = \inf\{d(x,a) : a \in A\}. \end{align*} [/definition] The nonempty hypothesis is part of the definition, not a harmless detail. If $A=\varnothing$, the displayed infimum has no real value under the usual convention, so this page only uses $\operatorname{dist}(\cdot,A)$ for nonempty sets. The question is whether this numerical device always produces the open neighbourhoods demanded by normality. For disjoint closed sets $A$ and $B$, a point cannot have distance zero from both sets, and comparing the two distance functions suggests a natural way to divide the space into an $A$-side and a $B$-side. In fact the metric argument gives more than ordinary normality. A space is called perfectly normal when it is normal and every closed subset is a $G_\delta$ set, meaning a countable intersection of open sets. Thus perfect normality strengthens normality by adding a countable open approximation to every closed set. [quotetheorem:1000] The theorem says that [metrizable spaces are perfectly normal](/theorems/1000), not merely normal. The examples below focus first on the normality part of this conclusion: the distance between two closed sets can be zero, but the distance-to-a-set functions still separate them by open neighbourhoods. The additional $G_\delta$ conclusion explains why metric closed sets can also be recovered from countably many open thickenings. [example: Disjoint Closed Sets with Zero Distance] Let $X=\mathbb{R}^2$ with its usual [Euclidean metric](/page/Euclidean%20Metric). Define \begin{align*} A=\{(x,0):x\in\mathbb{R}\},\qquad B=\{(x,e^x):x\in\mathbb{R}\}. \end{align*} The set $A$ is closed because it is the zero set of the continuous function $(x,y)\mapsto y$, and $B$ is closed because it is the zero set of the continuous function $(x,y)\mapsto y-e^x$. They are disjoint: if $(x,0)=(t,e^t)$, then $x=t$ and $0=e^t$, contradicting $e^t>0$. We compute the set-to-set distance. Since Euclidean distances are nonnegative, \begin{align*} 0\leq \inf\{|a-b|:a\in A,\ b\in B\}. \end{align*} For each $t\in\mathbb{R}$, take $a_t=(t,0)\in A$ and $b_t=(t,e^t)\in B$. Then \begin{align*} |a_t-b_t|=|(t,0)-(t,e^t)|=|(0,-e^t)|=\sqrt{0^2+(-e^t)^2}=e^t. \end{align*} Given any $\varepsilon>0$, choose $t<\log \varepsilon$. Then $e^t<\varepsilon$, so \begin{align*} \inf\{|a-b|:a\in A,\ b\in B\}\leq |a_t-b_t|=e^t<\varepsilon. \end{align*} Thus the infimum is nonnegative and smaller than every positive $\varepsilon$, hence \begin{align*} \inf\{|a-b|:a\in A,\ b\in B\}=0. \end{align*} Nevertheless, by *Metric Spaces Are Normal*, the disjoint closed sets $A$ and $B$ have disjoint open neighbourhoods in $\mathbb{R}^2$. The example shows why normality of metric spaces cannot be proved by assuming a positive global gap between closed sets; the separation must instead use the pointwise functions $\operatorname{dist}(x,A)$ and $\operatorname{dist}(x,B)$. [/example] This example is a useful warning. Normality is not the assertion that two closed sets have positive distance. It is the subtler assertion that the topology has enough room to separate them even when they approach each other at infinity. ## Shrinking Neighbourhoods Normality is often used in a form that looks stronger than the definition. Instead of merely placing a closed set inside an open set, we want to place it inside an open set whose closure still remains inside the prescribed open set. This is the operation of shrinking an open neighbourhood without losing the closed set. The next definition names the kind of controlled containment that appears in these arguments. [definition: Closed Shrinking] Let $(X,\tau)$ be a topological space, let $A \subset X$, and let $W \in \tau$ be an open neighbourhood of $A$. A closed shrinking of $W$ around $A$ is an open set $U \in \tau$ such that \begin{align*} A \subset U \subset \overline{U} \subset W. \end{align*} [/definition] The problem is to know when such a shrinking exists without a metric radius. If $A$ is closed and $W$ is open, then $X\setminus W$ is the forbidden closed set that the closure of the smaller neighbourhood must avoid. Normality is designed precisely to separate these two closed sets and should therefore produce the desired buffer. [quotetheorem:9167] This theorem is one of the main working forms of normality. It translates separation of two closed sets into a controlled buffer around one closed set inside a chosen open set. [example: Shrinking an Interval Neighbourhood] In $\mathbb{R}$ with its usual topology, let $A=[0,1]$ and $W=(-1,2)$. We show that $U=(-\tfrac12,\tfrac32)$ is a closed shrinking of $W$ around $A$. The set $U$ is open in the usual topology. If $x\in A$, then $0\leq x\leq 1$, so \begin{align*} -\tfrac12<0\leq x\leq 1<\tfrac32. \end{align*} Hence $x\in U$, and therefore $A\subset U$. The closure of $U$ is $[-\tfrac12,\tfrac32]$. Indeed, every point of $(-\tfrac12,\tfrac32)$ is already in $U$; the endpoint $-\tfrac12$ is a [limit point](/page/Limit%20Point) because for every $\varepsilon>0$, the point $-\tfrac12+\min(\varepsilon/2,1/2)$ lies in $U$; and the endpoint $\tfrac32$ is a limit point because $\tfrac32-\min(\varepsilon/2,1/2)$ lies in $U$. If $x<-\tfrac12$, then the open interval $(x-\delta,x+\delta)$ with $\delta=(-\tfrac12-x)/2$ misses $U$; if $x>\tfrac32$, then the open interval $(x-\delta,x+\delta)$ with $\delta=(x-\tfrac32)/2$ misses $U$. Thus no point outside $[-\tfrac12,\tfrac32]$ lies in the closure. Finally, \begin{align*} -\tfrac12>-1 \quad \text{and} \quad \tfrac32<2. \end{align*} So every $x\in[-\tfrac12,\tfrac32]$ satisfies $-1<x<2$, which gives $\overline{U}=[-\tfrac12,\tfrac32]\subset W$. Therefore \begin{align*} [0,1]\subset(-\tfrac12,\tfrac32)\subset[-\tfrac12,\tfrac32]\subset(-1,2). \end{align*} The smaller open interval surrounds $A$, and its closure still remains strictly inside the original neighbourhood $W$. [/example] Shrinking is the step that lets local definitions survive overlap. When building continuous functions or partitions of unity, one often first chooses large open sets where something is allowed to happen, and then shrinks to smaller sets where it actually happens. ## Continuous Separation Open-set separation is powerful, but analysis usually wants functions. A continuous function can encode a closed condition by taking one value on one set and another value elsewhere. Normality is exactly the topological hypothesis behind the most important version of this principle. Before stating that theorem, we isolate the target object. A function into $[0,1]$ is useful because the endpoints can mark two closed sets while the interior values describe a controlled transition between them. [definition: Separating Continuous Function] Let $(X,\tau)$ be a topological space, and let $A,B \subset X$ be disjoint subsets. A separating continuous function for $A$ and $B$ is a continuous map $f: X \to [0,1]$ such that $f(x)=0$ for all $x \in A$ and $f(x)=1$ for all $x \in B$. [/definition] Such a function is stronger than a pair of disjoint open neighbourhoods. Indeed, the inverse images of intervals near $0$ and $1$ give separated open sets, while the intermediate values describe a controlled transition region. The central question is whether the open-set separation in the definition of normality is strong enough to build such a function with no metric or coordinates. [quotetheorem:887] [Urysohn's lemma](/theorems/887) is the moment where normality becomes analytic. It converts a qualitative separation axiom into a continuous real-valued function, which can then be added, multiplied, truncated, and used in approximation arguments. [example: A Separating Function on the Real Line] Let $A=(-\infty,0]$ and $B=[1,\infty)$ in $\mathbb{R}$. These sets are disjoint because no real number can satisfy both $x\leq 0$ and $x\geq 1$. Define $f:\mathbb{R}\to[0,1]$ by setting $f(x)=0$ when $x\leq 0$, $f(x)=x$ when $0<x<1$, and $f(x)=1$ when $x\geq 1$. The values really lie in $[0,1]$: in the first and third regions the values are $0$ and $1$, and in the middle region $0<x<1$ gives $f(x)=x\in[0,1]$. For $x\in A$ we have $x\leq 0$, so the definition gives $f(x)=0$. For $x\in B$ we have $x\geq 1$, so the definition gives $f(x)=1$. It remains to check continuity at the joining points $0$ and $1$, since away from those points $f$ is either constant or the identity map. At $0$, let $\varepsilon>0$ and take $\delta=\min(\varepsilon,1)$. If $|x-0|<\delta$, then either $x\leq 0$, in which case \begin{align*} |f(x)-f(0)|=|0-0|=0<\varepsilon, \end{align*} or $0<x<1$, in which case \begin{align*} |f(x)-f(0)|=|x-0|=|x|<\delta\leq\varepsilon. \end{align*} Thus $f$ is continuous at $0$. At $1$, let $\varepsilon>0$ and take $\delta=\min(\varepsilon,1)$. If $|x-1|<\delta$, then $x>0$. If $0<x<1$, then \begin{align*} |f(x)-f(1)|=|x-1|<\delta\leq\varepsilon. \end{align*} If $x\geq 1$, then \begin{align*} |f(x)-f(1)|=|1-1|=0<\varepsilon. \end{align*} Thus $f$ is continuous at $1$, and therefore continuous on all of $\mathbb{R}$. This function separates the two closed regions by using the interval $(0,1)$ as the transition zone from value $0$ to value $1$. [/example] The real line example hides the hard part because the order gives a visible interpolation. In a general normal space there may be no coordinates, no metric, and no line segment joining the closed sets. The theorem says that normality alone still supplies enough nested open sets to manufacture the function. Once such functions exist, the next natural question is whether a function already defined on a closed subset can be extended to the whole space. [quotetheorem:888] Tietze's theorem says that in a normal space, closed subsets are not hostile domains for bounded continuous functions. A function defined on a closed part of the space can be extended to the whole space without breaking continuity. The closedness hypothesis is essential in spirit: values prescribed on a dense nonclosed subset can force a unique limiting behaviour, and an arbitrary continuous function there need not have compatible limits at boundary points. The bounded range is also part of the standard form quoted here, because it lets the extension be built while keeping the values controlled. This result turns Urysohn-type separation functions into a flexible tool: instead of only assigning $0$ and $1$ on two closed sets, one can prescribe a whole bounded continuous profile on a closed subspace and then continue it across the ambient normal space. ## Compactness and Local Compactness ### Compact Hausdorff Spaces Normality is not automatic for every topological space, but it follows from several familiar compactness hypotheses. Compact Hausdorff spaces are the central example: compactness turns open covers into finite data, while Hausdorffness separates points enough to build neighbourhoods of closed sets. To state this cleanly, we recall the compactness condition in the language of open covers. [definition: Compact Topological Space] A topological space $(X,\tau)$ is compact if every [open cover](/page/Open%20Cover) of $X$ has a [finite subcover](/page/Finite%20Subcover). [/definition] Compactness is a global finiteness principle. It does not itself say that points or closed sets can be separated, so the Hausdorff hypothesis remains essential. Together, compactness and Hausdorffness let one separate a closed set from each point of another closed set and then reduce infinitely many local separations to finitely many. [quotetheorem:1027] This theorem explains why normality appears throughout analysis on compact spaces. Once a space is compact Hausdorff, closed-set separation and Urysohn functions become available without an additional metric assumption. [example: The Circle Is Normal] Let $S^1=\{z\in\mathbb{C}:|z|=1\}$ with the [subspace topology](/page/Subspace%20Topology) inherited from $\mathbb{C}$. The map $z\mapsto |z|$ is continuous, and $\{1\}$ is closed in $\mathbb{R}$, so $S^1=\{z\in\mathbb{C}:|z|=1\}$ is closed in $\mathbb{C}$. Also, if $z\in S^1$, then $|z|=1$, so $S^1$ is bounded. By *Heine-Borel*, a closed and bounded subset of $\mathbb{C}\cong\mathbb{R}^2$ is compact; hence $S^1$ is compact. The space $S^1$ is Hausdorff because it is a subspace of the metric space $\mathbb{C}$. Explicitly, if $z,w\in S^1$ and $z\neq w$, set $r=|z-w|/3>0$. The sets $B(z,r)\cap S^1$ and $B(w,r)\cap S^1$ are open in $S^1$ and contain $z$ and $w$, respectively. If some $y$ belonged to both, then the triangle inequality would give \begin{align*} |z-w|\leq |z-y|+|y-w|<r+r=2|z-w|/3, \end{align*} which is impossible. Thus the two neighbourhoods are disjoint. Therefore $S^1$ is compact Hausdorff, so by *[Compact Hausdorff Spaces Are Normal](/theorems/1027)*, $S^1$ is normal. Consequently, whenever $A,B\subset S^1$ are disjoint closed subsets, there exist open sets $U,V$ in the subspace topology on $S^1$ such that \begin{align*} A\subset U,\quad B\subset V,\quad U\cap V=\varnothing. \end{align*} This applies equally to closed arcs, closed Cantor-type subsets of the circle, and any other pair of disjoint closed subsets; normality is separating the closed sets, not relying on their geometric simplicity. [/example] ### Local Compactness Compactness can also be local. Locally compact Hausdorff spaces need not be compact, but they often behave compactly near each point. This is enough to make many function-space constructions work, especially when combined with compact support. The next definition records the basic local compact control without building in a separate regularity condition. It asks that each point sit inside an open set which is itself contained in some compact set. [definition: Locally Compact Topological Space] A topological space $(X,\tau)$ is locally compact if for every point $x \in X$, there exist an open set $U \in \tau$ and a compact set $K \subset X$ such that \begin{align*} x \in U \subset K. \end{align*} [/definition] This version of local compactness says only that compact sets are available near each point. In Hausdorff settings, compact sets are closed, and compact neighbourhoods can be used to localize continuous functions. The analytic payoff is a bump-function principle: when a compact set sits inside an open set, one wants a continuous function that is identically $1$ on the compact set and vanishes outside the chosen open set, so that all activity is confined to a controlled region. The remaining issue is whether local compact control is strong enough to build such functions without assuming a global normality theorem in advance. The following result answers that local extension question for compact subsets inside open neighbourhoods, giving exactly the support control needed in analysis. Here $C_c(X)$ denotes the space of continuous scalar-valued functions on $X$ whose support is compact, meaning the closure of the set on which the function is nonzero is compact. [quotetheorem:1065] This compactly supported Urysohn lemma is stronger than mere point-closed separation in the direction needed for analysis. It does not assert that every locally compact [Hausdorff space](/page/Hausdorff%20Space) is normal; rather, it supplies local functions supported inside prescribed open neighbourhoods of compact sets. To obtain global normality, one needs additional hypotheses that let these local compact controls be organized across the whole space. [quotetheorem:9168] This result covers many spaces used in analysis, including open subsets of $\mathbb{R}^n$, manifolds with countable atlases, and many phase spaces. It also points toward a theme: normality is often obtained from a mixture of separation, compactness, and countability. ## Heredity, Products, and Failure Modes ### Subspaces A property is easier to use when it passes to subspaces and products. Normality behaves well under some operations and badly under others, which is why it must be handled with care. Closed subspaces are the most dependable hereditary case. To formulate this, we first recall the topology that a subset inherits from its ambient space. [definition: Subspace Topology] Let $(X,\tau)$ be a topological space and let $Y \subset X$. The subspace topology on $Y$ is \begin{align*} \tau_Y = \{U \cap Y : U \in \tau\}. \end{align*} [/definition] The subspace topology is the topology that makes inclusion $Y \hookrightarrow X$ continuous while preserving the open sets visible from $X$. The next issue is whether normality survives this passage to a subset. When $Y$ is closed in $X$, closed subsets of $Y$ are also obtained by intersecting $Y$ with closed subsets of $X$, so the ambient normality has exactly the information needed to separate them. [quotetheorem:9169] This permanence property is frequently used without comment: closed subsets of compact Hausdorff spaces, closed subsets of metric spaces, and closed subsets of familiar function spaces inherit normality. [example: A Closed Annulus] Let $Y=\{x\in\mathbb{R}^2:1\leq |x|\leq 2\}$, where $|x|$ is the Euclidean norm. We first check that $Y$ is closed. If $z\notin Y$, then either $|z|<1$ or $|z|>2$. When $|z|<1$, set $r=(1-|z|)/2>0$. If $|w-z|<r$, then by the triangle inequality, \begin{align*} |w|\leq |z|+|w-z|<|z|+\frac{1-|z|}{2}=\frac{1+|z|}{2}<1. \end{align*} So $w\notin Y$. When $|z|>2$, set $r=(|z|-2)/2>0$. If $|w-z|<r$, then by the [reverse triangle inequality](/theorems/2300), \begin{align*} |w|\geq |z|-|w-z|>|z|-\frac{|z|-2}{2}=\frac{|z|+2}{2}>2. \end{align*} So again $w\notin Y$. Thus every point of $\mathbb{R}^2\setminus Y$ has an open ball contained in $\mathbb{R}^2\setminus Y$, so $\mathbb{R}^2\setminus Y$ is open and $Y$ is closed. The space $\mathbb{R}^2$ is normal by *Metric Spaces Are Normal*, and $Y$ is a closed subspace of $\mathbb{R}^2$, so $Y$ is normal by *[Closed Subspaces of Normal Spaces Are Normal](/theorems/9169)*. For example, let \begin{align*} C_1=\{x\in\mathbb{R}^2:|x|=1\} \end{align*} and \begin{align*} C_2=\{x\in\mathbb{R}^2:|x|=2\}. \end{align*} Both $C_1$ and $C_2$ are subsets of $Y$. They are disjoint because no vector $x$ can satisfy both $|x|=1$ and $|x|=2$. Also, $C_1$ and $C_2$ are closed in $Y$: if $x\in Y\setminus C_1$, then $1<|x|\leq 2$, and the same ball argument with radius $(|x|-1)/2$ keeps nearby points away from $C_1$; if $x\in Y\setminus C_2$, then $1\leq |x|<2$, and the same ball argument with radius $(2-|x|)/2$ keeps nearby points away from $C_2$. Since $Y$ is normal, there exist open sets $U,V$ in the subspace topology on $Y$ such that \begin{align*} C_1\subset U,\quad C_2\subset V,\quad U\cap V=\varnothing. \end{align*} Thus the annulus inherits enough separation from the plane to give disjoint open neighbourhoods, inside the annulus itself, around its two boundary circles. [/example] The pleasant behaviour of closed subspaces does not extend to arbitrary subspaces. A subspace may contain closed sets that are not controlled by closed sets in the original space, so the next theorem marks a genuine boundary of the theory rather than a technical inconvenience. [quotetheorem:9170] The failure of arbitrary heredity is one reason normality is a delicate separation axiom. It is strong enough for analysis, but not so rigid that it behaves like metrisability under every construction. ### Products Products are even more delicate. Finite products of metric spaces are normal because they are metric, and products of compact Hausdorff spaces are compact Hausdorff. Outside those protected cases, closed subsets of a product can encode interactions between coordinates that neither factor sees alone. [quotetheorem:9171] This theorem prevents a common false inference. Normality is not merely a local property checked factor by factor; the [product topology](/page/Product%20Topology) can create new closed sets whose separation requires more than the factors provide. ## Role in Analysis Normal spaces enter analysis through the same door as cutoff functions: we need continuous functions that are prescribed on closed sets and flexible elsewhere. The topological theorem supplies the function; analytic estimates then use it. A support condition is one of the main ways this happens. The next definition is standard for functions on a topological space, and it records where a function is allowed to be nonzero. [definition: Support of a Continuous Function] Let $(X,\tau)$ be a topological space, and let $f: X \to \mathbb{R}$ be continuous. The support of $f$ is \begin{align*} \operatorname{supp} f = \overline{\{x \in X : f(x) \neq 0\}}. \end{align*} [/definition] Support connects topology to analysis because it turns open-set localisation into a statement about functions. The next problem is to build a function that equals $1$ on a closed set $K$ but has no effect outside a chosen open region $U$. Normality supplies this by separating $K$ from the closed complement $X\setminus U$ and then applying Urysohn's lemma. [quotetheorem:887] The statement is phrased to emphasize the mechanism. Normality separates the closed set $K$ from the closed complement $X\setminus U$, and the resulting continuous function vanishes outside $U$. This does not by itself say that $\operatorname{supp} f \subset U$; support is a closed set, so boundary points of $U$ may still lie in the support. [example: A Cutoff Function on $\mathbb{R}$] Let $K=[-1,1]$ and $U=(-2,2)$ in $\mathbb{R}$. Define $f:\mathbb{R}\to[0,1]$ by setting $f(x)=0$ when $|x|\geq 2$, $f(x)=2-|x|$ when $1<|x|<2$, and $f(x)=1$ when $|x|\leq 1$. The values lie in $[0,1]$: in the outer region the value is $0$, in the inner region the value is $1$, and if $1<|x|<2$, then \begin{align*} 0<2-|x|<1. \end{align*} The formula is continuous at the joining points. At $x=1$, the left-hand formula gives $1$, and the middle formula gives \begin{align*} 2-|1|=1. \end{align*} At $x=-1$, the left-hand formula gives $1$, and the middle formula gives \begin{align*} 2-|-1|=1. \end{align*} At $x=2$, the middle formula gives \begin{align*} 2-|2|=0, \end{align*} which matches the outer value. At $x=-2$, the middle formula gives \begin{align*} 2-|-2|=0, \end{align*} which again matches the outer value. Away from these four points, $f$ is locally constant or equal to $2-|x|$, so $f$ is continuous on $\mathbb{R}$. For every $x\in K$, we have $|x|\leq 1$, so $f(x)=1$. Also, \begin{align*} \mathbb{R}\setminus U=(-\infty,-2]\cup[2,\infty)=\{x\in\mathbb{R}:|x|\geq 2\}. \end{align*} Hence $f(x)=0$ for every $x\in\mathbb{R}\setminus U$. The nonzero set is exactly $(-2,2)$. If $|x|\geq 2$, then $f(x)=0$. If $|x|<2$, then either $|x|\leq 1$, giving $f(x)=1\neq 0$, or $1<|x|<2$, giving $f(x)=2-|x|>0$. Therefore \begin{align*} \{x\in\mathbb{R}:f(x)\neq 0\}=(-2,2). \end{align*} By the definition of support, \begin{align*} \operatorname{supp} f=\overline{(-2,2)}=[-2,2]. \end{align*} The function is a cutoff that vanishes outside $U$, but its support is not contained in $U$ because $-2\in\operatorname{supp} f$ and $2\in\operatorname{supp} f$, while neither endpoint belongs to $U=(-2,2)$. Thus vanishing outside an open set is weaker than having support contained in that open set. [/example] In smooth analysis one asks for smooth cutoff functions, which require differentiable structure and stronger tools such as mollification. Normality is the purely continuous precursor: it supplies continuous cutoffs before derivatives enter the story. ## Beyond and Connected Topics Normality sits between general topology and the function-building techniques used in analysis. The immediate parent concept is [topological space](/page/Topological%20Space), where open sets, closed sets, closure, and continuity are defined without separation assumptions. Normality adds a closed-set separation principle to that parent structure. Metric spaces provide the main source of examples. The theorem that metric spaces are normal explains why students first encounter normal behaviour in [Cambridge IB Analysis and Topology](/page/Cambridge%20IB%20Analysis%20and%20Topology), even if the abstraction is not initially named. In metric settings, distance functions replace the abstract nested-open-set construction. Compact Hausdorff spaces form a second source. They are central in functional analysis because compactness turns local open-cover data into finite data, and normality then supplies continuous functions. This theme continues naturally into [Cambridge II Linear Analysis](/page/Cambridge%20II%20Linear%20Analysis), where spaces of continuous functions and duality become central objects. Complex analysis uses normality less explicitly, but it relies constantly on separation by neighbourhoods and continuous auxiliary functions on subsets of $\mathbb{C}$. The metric normality of plane domains underlies many constructions in [Cambridge IB Complex Analysis](/page/Cambridge%20IB%20Complex%20Analysis), even when holomorphic tools later replace purely topological ones. Analysis of functions brings the same idea into approximation and regularity. The transition from continuous cutoffs to smooth cutoffs belongs to [Cambridge II Analysis of Functions](/page/Cambridge%20II%20Analysis%20of%20Functions), where topology, measure, and differentiability interact more tightly. ## References Androma, [Cambridge IB Analysis and Topology](/page/Cambridge%20IB%20Analysis%20and%20Topology). Androma, [Cambridge IB Complex Analysis](/page/Cambridge%20IB%20Complex%20Analysis). Androma, [Cambridge II Analysis of Functions](/page/Cambridge%20II%20Analysis%20of%20Functions). Androma, [Cambridge II Linear Analysis](/page/Cambridge%20II%20Linear%20Analysis). James R. Munkres, *Topology* (2000). John L. Kelley, *General Topology* (1955). Stephen Willard, *General Topology* (1970).