Disjoint Closed Sets and Normality in Topology

In topology, the intuitive notion of two objects being "separate" requires a surprisingly precise and powerful framework. While simple disjointness isn't enough, the ability to robustly separate two closed sets that do not intersect proves to be a fundamental property that distinguishes different types of topological spaces. This article addresses the gap between our intuitive understanding of separation and its rigorous mathematical formulation, revealing why this critical property is not always guaranteed.

The reader will embark on a journey through this concept, beginning with its core principles. The "Principles and Mechanisms" chapter will define normal spaces and introduce Urysohn's Lemma, a landmark theorem that forges a deep connection between the geometry of separation and the analysis of continuous functions. Following this, the "Applications and Interdisciplinary Connections" chapter will explore where our intuition holds true, where it dramatically fails through a gallery of fascinating counterexamples, and how the property of normality becomes a gateway to advanced topics like dimension theory. By examining both the elegant theory and its perplexing limits, we will gain a deeper appreciation for the rich structure governing the nature of space.

In our journey through the world of shapes and spaces, some of the most profound ideas arise from the simplest questions. Let’s start with one: what does it mean for two objects to be separate? You might say it's easy—they don't touch. Their intersection is empty. But in the fluid, stretchy world of topology, "not touching" can be a surprisingly slippery concept. This subtlety is not just a mathematical curiosity; it is the gateway to understanding a deep and beautiful structure that governs the very nature of space.

Imagine two open intervals on the real number line, say $A = (0, 1)$ and $B = (2, 3)$. They are disjoint, and there's a comfortable gap of one whole unit between them. Now consider a different pair: $A = (0, 1)$ and $B = (1, 2)$. They are still disjoint—the number 1 belongs to neither—but they feel uncomfortably close, "kissing" at that single point. This difference, which our intuition readily grasps, needs a more precise language in topology.

To formalize this, we use the concept of a set's closure. The ​​closure​​ of a set, denoted $\overline{A}$, is the set itself plus all of its limit points. Two sets $A$ and $B$ are called ​​separated​​ if neither set intersects the closure of the other. Formally, $\overline{A} \cap B = \emptyset$ and $A \cap \overline{B} = \emptyset$. Our "kissing" intervals, $A=(0,1)$ and $B=(1,2)$, are indeed separated. While their closures do touch ($\overline{A} \cap \overline{B} = \{1\}$), the sets themselves manage to stay out of each other's extended territory. So, being "separated" is a stricter condition than just being disjoint.

However, the most robust form of separation, the one that truly sets the stage for deeper results, involves sets that already contain all their boundary points. These are the ​​closed sets​​. Consider two disjoint closed intervals, like $A = [0, 1]$ and $B = [2, 3]$. They are disjoint, and because they are closed, they are automatically separated from each other. There is a palpable "no man's land" between them. It is this situation—two ​​disjoint closed sets​​—that turns out to be the key that unlocks a new level of order in a topological space.

If we have two disjoint closed sets, how can we be sure that the space itself acknowledges this separation? We need a guarantee that we can build a wall, or a moat, around each set without the walls touching. This guarantee is the defining characteristic of a ​​normal space​​.

A topological space $X$ is called ​​normal​​ if, for any two disjoint closed sets $A$ and $B$, you can always find two disjoint open sets, $U$ and $V$, that contain them. That is, $A \subseteq U$, $B \subseteq V$, and $U \cap V = \emptyset$. (For technical precision, we also require the space to be a ​​T1 space​​, where individual points are closed sets, a property held by most familiar spaces).

Imagine $A$ and $B$ are two islands. Normality is the promise that you can always dredge a channel ($U$) around island $A$ and a separate channel ($V$) around island $B$, and these two bodies of water will never merge. This seems like a reasonable property, but not all spaces have it. Consider a strange little universe with just four points, $X = \{a, b, c, d\}$, and a specific collection of "open" sets. It's possible to design this space such that $\{c\}$ and $\{d\}$ are both closed sets, yet every open set that contains $c$ inevitably overlaps with every open set that contains $d$. In such a world, our islands are so strangely intertwined that it's impossible to build separate moats. This space is not normal.

The property of normality is a statement of tidiness and order. It is a powerful condition. For instance, any normal space is automatically ​​regular​​, meaning you can always separate a point from a closed set that doesn't contain it. This makes sense: since points are just very small closed sets in a T1 space, this is simply a special case of normality. Normality sits high in the hierarchy of separation axioms, bringing with it a suite of powerful consequences. One of them is so profound it deserves a section all to itself.

So, a normal space allows us to put disjoint open sets around disjoint closed sets. This is a nice, qualitative geometric property. But where's the magic? The magic lies in a stunning result by the mathematician Pavel Urysohn, which builds a bridge from this abstract property to the world of continuous functions.

​​Urysohn's Lemma​​ states that if a space $X$ is normal, then for any two disjoint closed sets $A$ and $B$, there exists a ​​continuous function​​ $f: X \to [0, 1]$ such that $f(x)=0$ for all $x$ in $A$, and $f(x)=1$ for all $x$ in $B$.

Think about what this means. The simple rule about separating islands with moats guarantees that we can "paint" the entire space with a continuous gradient of colors, from pure black (value 0) on island $A$ to pure white (value 1) on island $B$. The function $f$ acts as a topological landscape, creating a smooth ramp that starts at elevation 0 on $A$ and rises steadily to 1 on $B$. This is not just some function; its very existence is a deep feature of the space's structure.

How is such a miracle possible? The proof is a beautiful display of ingenuity, like building a staircase across a chasm, step by step.

1. We start with our disjoint closed sets $A$ and $B$. Normality doesn't just give us one pair of open sets; it gives us something more powerful. It implies that if you have a closed set $F$ inside an open set $W$, you can always find another open set $U$ to squeeze in between: $F \subseteq U \subseteq \overline{U} \subseteq W$. This is like being able to build a slightly smaller moat inside a larger one, with a stone wall (the closure) separating them.

2. Let's use this. Set $A$ is our closed set, and $U_1 = X \setminus B$ is our open set containing it. We can squeeze an open set $U_0$ in between: $A \subseteq U_0 \subseteq \overline{U_0} \subseteq U_1$. We have now bracketed $A$ very tightly.

3. Now for the magic. We have a closed set $\overline{U_0}$ inside an open set $U_1$. They are perfect candidates for our squeezing property again! We can find a new open set, let's call it $U_{1/2}$, such that $\overline{U_0} \subseteq U_{1/2} \subseteq \overline{U_{1/2}} \subseteq U_1$. We've just built a step halfway between 0 and 1!

4. There's nothing special about $1/2$. We can repeat this process for all the dyadic rationals (numbers like $1/4, 3/4, 1/8, 3/8, \dots$). This creates a whole family of nested open sets $\{U_r\}$ indexed by these rationals $r \in [0,1]$, where if $r s$, then $\overline{U_r} \subseteq U_s$.

5. With this ladder of open sets in place, we can define our function. For any point $x$ in our space, we simply ask: what is the first rung of the ladder that $x$ steps onto? We define $f(x)$ to be the smallest (infimum) of all rational indices $r$ such that $x \in U_r$. This clever definition produces the desired continuous function that is 0 on $A$ and 1 on $B$.

From a simple rule about sets, we have conjured a sophisticated analytical object. This is the beauty and unity of mathematics.

This "Urysohn function" is more than a party trick; it's a powerful tool. For instance, it provides an elegant proof that every normal space is also ​​completely regular​​ (or a ​​Tychonoff space​​). A space is completely regular if you can separate any point $p$ from a closed set $C$ not containing it with a continuous function. How do we prove this? Easy. In a normal (and T1) space, the point $\{p\}$ is itself a closed set. So we have two disjoint closed sets, $\{p\}$ and $C$. Apply Urysohn's Lemma directly, and out pops a function that is 0 at $p$ and 1 on $C$. The lemma does all the work.

The specific target interval $[0,1]$ isn't sacred. A simple transformation like $g(x) = 2f(x)-1$ gives a continuous function to $[-1,1]$ that separates the sets with values $-1$ and $1$. The existence of a separating function to $[0,1]$ is, in fact, equivalent to the existence of one to any closed interval $[a,b]$.

But we must be careful with our generalizations. What if we wanted a function to a disconnected space, like the two-point set $\{-1, 1\}$? This would mean the space $X$ itself could be partitioned into two disjoint open sets, one containing $A$ and the other containing $B$. This is a much stronger condition than normality and is not guaranteed by it. A continuous function cannot create a disconnection where none exists; the continuous image of a connected space (like the real line $\mathbb{R}$) must be connected.

Finally, does a Urysohn function $f$ only separate the original sets $A$ and $B$? Not necessarily. By its very construction, the function $f$ takes the value 0 on the entire set $A_0 = f^{-1}(\{0\})$ and 1 on $B_0 = f^{-1}(\{1\})$. If $f$ also separates another pair of sets $(A', B')$, it must be that $A' \subseteq A_0$ and $B' \subseteq B_0$. The function is defined by the largest possible sets it can separate.

Urysohn's Lemma is so powerful, it's tempting to think it can do anything. Let's push it. If it can handle two sets, what about a countably infinite collection of disjoint closed sets, $\{A_n\}_{n=1}^{\infty}$? Can we always find a single continuous function $f$ that neatly tags each set, say with the value $f(x) = 1/n$ for all $x \in A_n$?

The answer, perhaps surprisingly, is no. The magic has its limits, and the reason reveals something deep about the nature of continuity.

Imagine a scenario where a point $p$ in one set, say $A_k$, is a limit point for a sequence of points $\{p_m\}$ taken from other sets. Let's say $p_m \in A_{j_m}$, where the indices $j_m \to \infty$ as $m \to \infty$. Now, suppose our grand separating function $f$ exists. By the very definition of continuity, if $p_m \to p$, then we must have $f(p_m) \to f(p)$. Let's check the values.

• On the one hand, $f(p_m) = 1/j_m$. Since $j_m \to \infty$, this means the sequence of function values $f(p_m)$ converges to 0.
• On the other hand, $f(p)$ must be $1/k$, since $p \in A_k$.

Continuity demands that these limits be the same: $f(p)$ must equal the limit of $f(p_m)$. This forces $1/k = 0$, which is impossible. The function we wish to construct is torn apart by the demands of continuity at this limit point. The values we want to assign ($1/n$) cluster toward 0, and if the points in the corresponding sets ($A_n$) cluster toward a point in some other set ($A_k$), the function cannot remain continuous.

This beautiful failure teaches us a crucial lesson. The power of normality, as expressed by Urysohn's Lemma and its cousin the Tietze Extension Theorem, is fundamentally about separating a finite number of closed sets, or more precisely, separating two closed sets from each other. When we move to an infinite collection, new and subtle topological behaviors, like the clustering of limit points, can arise and prevent our simple, elegant constructions from working. And in discovering these limits, we gain an even deeper appreciation for the delicate and powerful machinery that holds the world of topology together.

In our journey through the world of topology, we often begin with ideas that feel deeply familiar, rooted in our everyday experience of space. We can tell when two objects are separate. We can imagine drawing a boundary between two countries, or putting a fence between two yards. This intuitive act of separation is the heart of what we are about to explore. But as we shall see, when we try to make this simple idea precise, we are led down a path of surprising discoveries, where our intuition is both a guide and something to be challenged. We will see that the seemingly simple question, "Can we always separate two disjoint closed things?", opens a door to a veritable zoo of strange and beautiful mathematical structures, connecting topology to analysis, dimension theory, and the very nature of continuity itself.

Let's begin in comfortable territory. The first and most basic kind of separation we learn in topology is the ability to separate two distinct points with two disjoint open sets—a property called the Hausdorff condition. This is like saying that any two distinct specks of dust in a room can be enclosed in their own separate, non-overlapping soap bubbles. This feels right. It's a feature of any "reasonable" space.

What if we want to separate not just points, but larger objects? Suppose we have two disjoint sets that are also compact. In topology, compactness is a powerful notion of "finiteness" or "boundedness." Think of a closed and bounded shape in the familiar Euclidean plane, like a disk or a square. Can we always find two disjoint open "neighborhoods" that contain them?

Happily, in any Hausdorff space, the answer is a resounding yes! The logic is as beautiful as it is powerful. You can imagine taking a single point in the first compact set and, using the Hausdorff property, separating it from every single point in the second set. This gives you an open cover for the second set. Because the second set is compact, you only need a finite number of these open sets to cover it. By carefully taking intersections and unions, you build an open bubble around your original point that is completely disjoint from an open bubble containing the entire second set. You then repeat this for every point in the first set, and once again use compactness to select a finite number of bubbles to construct your final separating sets. It's a marvelous bootstrapping process, building from the separation of points to the separation of entire compact regions. Our intuition holds up perfectly.

The next logical step is to ask: can we do this for any two disjoint closed sets? A closed set is one that contains all of its boundary points. While every compact set in a Hausdorff space is closed, not all closed sets are compact (think of an infinite line in a plane). This generalization from "compact" to "closed" seems minor, but it is a chasm. A space where any two disjoint closed sets can be separated by disjoint open sets is called a ​​normal space​​.

The normality property has a stunningly beautiful equivalent formulation, given by the famous Urysohn's Lemma. This theorem states that a space is normal if and only if for any two disjoint closed sets, say $A$ and $B$, you can define a continuous function on the entire space that takes the value $0$ on all of $A$ and the value $1$ on all of $B$. It's like building a smooth landscape over your space that is at sea level ($0$) on the continent of $A$ and rises to a uniform plateau of height $1$ on the continent of $B$. The existence of this "separating landscape" is geometrically and analytically a far more powerful and useful idea than just finding two open sets. It connects the topological property of separation to the world of analysis and continuous functions. In the familiar spaces we know, like the real line or Euclidean space, this is always possible. Our intuition, trained on these examples, screams that this must always be true.

And this is where our comfortable journey takes a sharp turn into the wilderness.

It turns out that normality is not a given. Many spaces, even those constructed from simple pieces, fail this crucial test in fascinating ways. Let's tour a gallery of these "pathological" but deeply instructive spaces.

​​The Line with Two Origins:​​ Imagine taking the real line and splitting the point zero into two distinct points, let's call them $p_1$ and $p_2$. We define the topology such that any open set containing $p_1$ must contain a small interval $(-\epsilon, \epsilon)$ (excluding zero), and similarly for $p_2$. The single-point sets $\{p_1\}$ and $\{p_2\}$ are closed and disjoint. But can we separate them? No! Any open "bubble" around $p_1$ and any open "bubble" around $p_2$ are forced to share the same punctured intervals around the original zero. They will always overlap. Our attempt to separate two points created a space where those very points, now closed sets, have become inseparable.

​​The K-Topology:​​ Here we take the ordinary real line and just slightly alter its open sets. The usual open intervals are still open, but we also allow sets of the form $(a,b) \setminus K$, where $K$ is the set of points $\{1, 1/2, 1/3, \dots\}$. Now consider the set $K$ itself and the point $\{0\}$. Both are closed sets, and they are disjoint. But any attempt to put an open bubble around $K$ requires us to include small open intervals around each point $1/n$. As $n$ gets large, these intervals get arbitrarily close to $0$. Any open bubble we try to draw around $0$ will inevitably be pierced by one of these intervals. The set $K$ "ambushes" the origin, making separation impossible.

​​The Geometric Counterexamples:​​ More complex failures of normality arise in spaces that look, at a glance, like the ordinary plane.

• The ​​Sorgenfrey plane​​ is built by taking the product of two real lines, each with a topology where basic open sets are half-open intervals $[a, b)$. This subtle change has dramatic consequences. Consider the line $y = -x$. The set of points on this line with rational coordinates, $A$, and the set of points with irrational coordinates, $B$, are both closed sets in this strange topology. They are clearly disjoint. Yet they cannot be separated. The proof reveals that any open set containing the irrational points must be "thick" enough that the dense set of rational points can always find a way to poke through into it.
• The ​​Moore plane​​ (or Niemytzki plane) is the upper half-plane with a special topology on the x-axis, where open sets containing a boundary point $(x,0)$ look like open disks tangent to the axis at that point. Again, the set of rational points on the x-axis and the set of irrational points are disjoint closed sets. Proving they cannot be separated leads to a magnificent connection with analysis: the existence of a separating Urysohn function for these sets would imply the existence of a sequence of continuous functions converging pointwise to the everywhere-discontinuous Dirichlet function, a well-known impossibility. The topological problem of separation is, in this case, identical to a classic problem in real analysis.

​​Failures of Scale:​​ Non-normality can also arise from issues of "size" and "dimension."

• In an uncountable set with the ​​cocountable topology​​ (where open sets have countable complements), the open sets are simply "too big." Any two non-empty open sets must intersect. This makes it impossible to find disjoint open sets to separate anything at all, such as two distinct points.
• In the ​​infinite product of real lines with the box topology​​, $\mathbb{R}^\omega$, the failure comes from having too much freedom. A basic open set can be a product of open intervals where every interval in the infinite product can be made arbitrarily small, with no coordination between them. This allows for the construction of disjoint closed sets whose open neighborhoods can be made so "thin" and "pointy" in infinitely many dimensions that they are forced to intersect. It's a failure of coordination in infinite dimensions.

This tour of counterexamples might leave one feeling that topology is just a collection of monsters. But each monster teaches us a lesson. One of the most important is the need for precision. One might be tempted to ask, "Well, can we separate the rationals and irrationals on the simple Sorgenfrey line?" But here we must pause. The question of normality is about separating disjoint closed sets. In the Sorgenfrey line, neither the set of rationals nor the set of irrationals is closed! Their closures are the entire line. The game of separation has rules, and the first rule is that the sets you're trying to separate must be closed.

So, normality is a special, non-universal property. But where it exists, it is powerful. It is not just a static label; it is a property that can be passed on. If you take a normal space $X$ and map it onto another space $Y$ with a function that is continuous, closed, and surjective (meaning it covers all of $Y$), then the space $Y$ is guaranteed to be normal as well. The property survives the journey.

Perhaps the most profound application of normality is as a gateway to other deep concepts. Consider the idea of dimension. How can we define what it means for an abstract space to be $n$-dimensional? One way, the ​​large inductive dimension​​, does so recursively. It says, roughly, that a space has dimension at most $n$ if any two disjoint closed sets can be separated by a "wall" (another closed set) whose dimension is at most $n-1$. But look at the very first part of that condition: "if for any two disjoint closed sets...". This is exactly the definition of normality! If a space is not normal, it fails this condition at the most basic level. There exists at least one pair of disjoint closed sets that cannot be separated at all, let alone by a lower-dimensional wall. For such a space, like the Sorgenfrey plane, the entire dimension theory program cannot even get started. Its large inductive dimension is declared to be infinite, not because it is complex in a high-dimensional way, but because it fails the fundamental entry requirement of normality.

From a simple geometric question about drawing boundaries, we have journeyed through the nature of continuity, uncovered a zoo of bizarre spaces, and found a critical prerequisite for the theory of dimension. The study of separating disjoint closed sets is a perfect example of the mathematical process: we follow our intuition until it breaks, and in studying the pieces, we discover a richer, deeper, and more beautiful structure than we ever imagined.