Counterexamples in Topology

In the abstract world of topology, where familiar notions of distance and shape are replaced by the raw essence of structure, our intuition can be a treacherous guide. How do we build a solid understanding of concepts like "connectedness" or "separation" when the rules are so flexible? The answer lies not just in theorems that work, but in understanding precisely when and why they fail. This article addresses this challenge by focusing on one of topology's most powerful learning tools: the counterexample. These are not mathematical errors, but masterfully constructed spaces that serve as stress tests for our ideas, revealing hidden assumptions and the true boundaries of our theories.

This article is structured to guide you through this fascinating landscape of "beautiful monsters." In the first chapter, ​​"Principles and Mechanisms,"​​ we will delve into the foundational axioms of a topological space and explore the hierarchy of separation axioms, using counterexamples to dissect why these rules are crafted so precisely. In the following chapter, ​​"Applications and Interdisciplinary Connections,"​​ we will visit a gallery of famous topological counterexamples, such as the Topologist's Sine Curve and the Sorgenfrey plane, to see how they challenge our assumptions and deepen our appreciation for the rigor and beauty of the subject.

In our journey into topology, we've seen that it is the study of shapes and spaces where the rigid notions of distance and angle are thrown away. What's left is the very essence of "connectedness," "nearness," and "insideness." But how do we build a world with such flexible rules and still have it make any sense at all? The answer lies in a simple, yet profound, set of axioms that define what we call a ​​topological space​​. These axioms are not arbitrary; they are the bedrock upon which all of topology is built. Like the rules of a game, they are spare, but their consequences are vast and often surprising. It is by pushing these rules to their limits, by asking "what if...?" and constructing strange new worlds—our counterexamples—that we truly begin to understand the game.

Let's start at the beginning. A topology on a set $X$ is simply a collection $\mathcal{T}$ of subsets of $X$, which we call the ​​open sets​​. This collection must obey three rules:

1. The empty set, $\emptyset$, and the entire set, $X$, must be open.
2. The union of any number of open sets (finite, infinite, or even uncountably infinite) must also be open.
3. The intersection of a finite number of open sets must also be open.

The first rule is for bookkeeping. The second rule says we can glue open regions together and get another open region. It’s the third rule that often feels a bit peculiar. Why only finite intersections? Why not arbitrary intersections?

Let’s play a game. Imagine our set is the natural numbers, $\mathbb{N} = \{1, 2, 3, \dots\}$. What if we tried to define a "topology" where the open sets are all the infinite subsets of $\mathbb{N}$, plus $\emptyset$ and $\mathbb{N}$ itself? This seems plausible; "open sets" feel like they should be large. But this collection fails to be a topology, and the reason reveals the wisdom of the third axiom.

Consider the set of all multiples of 3, $U_1 = \{3, 6, 9, \dots\}$, and another set $U_2 = \{3\} \cup \{n \in \mathbb{N} \mid n \equiv 1 \pmod 3\}$. Both $U_1$ and $U_2$ are clearly infinite, so they are "open" in our proposed collection. But what is their intersection? An element must be a multiple of 3 and also either be the number 3 or be one more than a multiple of 3. The only number that satisfies this is 3 itself. So, $U_1 \cap U_2 = \{3\}$. This is a finite, non-empty set. According to our proposed rules, it is not open. We took two "open" sets, found their intersection, and landed outside our collection. The structure collapsed. This is precisely the kind of instability the third axiom prevents. The axioms aren't just a list; they are a carefully engineered system to ensure that the basic operations of union and intersection don't unexpectedly kick you out of the world you've defined.

This distinction also highlights a fundamental difference between the world of topology and the world of measure theory. A collection of sets for measuring things, a ​​$\sigma$-algebra​​, must be closed under complements and countable unions. A topology, on the other hand, is closed under arbitrary unions and finite intersections. This means an open set's complement, a ​​closed set​​, is not guaranteed to be open. Think of the open interval $(0, 1)$ on the real number line. It's a quintessential open set. Its complement, $(-\infty, 0] \cup [1, \infty)$, contains its boundary points and is not open. This seemingly simple difference—closure under complements—is a deep fork in the road, leading topology and measure theory down related but distinct paths.

Now that we have a space, what can we do with it? The first, most basic question we can ask is: if we have two different points, can our topology tell them apart? This simple question launches one of the great stories in topology: the ​​separation axioms​​. They form a hierarchy of properties, each describing a more refined ability to isolate points and sets from one another.

The most intuitive level of separation is what we call ​​Hausdorff​​, or $T_2$. It says that for any two distinct points $x$ and $y$, you can find two non-overlapping open sets, one containing $x$ and the other containing $y$. Think of putting a small bubble around each point so that the bubbles don't touch. This property is so fundamental that many mathematicians won't even consider a space a "space" unless it's Hausdorff.

But what if a space fails to be Hausdorff? Let's consider an infinite set, like the real numbers $\mathbb{R}$, but with a bizarre topology called the ​​cofinite topology​​. Here, a set is open if it's either empty or its complement is finite. In this world, every singleton set $\{x\}$ is closed, since its complement $\mathbb{R} \setminus \{x\}$ is cofinite and thus open. This property is called $T_1$. So we can distinguish individual points. But now try to find two disjoint, non-empty open sets. Let $U$ and $V$ be two such sets. By definition, their complements, $\mathbb{R} \setminus U$ and $\mathbb{R} \setminus V$, are both finite. The complement of their intersection is $(\mathbb{R} \setminus U) \cup (\mathbb{R} \setminus V)$, which is also a finite set. This means $U \cap V$ cannot be empty—it must be an infinite set! In the cofinite world, any two non-empty open sets must overlap. It's a space where you can isolate individual points, but you can never put non-overlapping bubbles around any two of them. This space is $T_1$ but spectacularly not Hausdorff, and because it isn't Hausdorff, it also fails to be ​​regular​​.

Speaking of regularity, let's climb the hierarchy. A space is ​​regular​​ if you can separate any point from a closed set that doesn't contain it. This seems stronger than Hausdorff. And a space is ​​normal​​ if you can separate any two disjoint closed sets. This seems stronger still. One might naturally assume this is a neat, linear progression: Normal $\implies$ Regular $\implies$ Hausdorff $\implies T_1$.

But here, a tiny, three-point set can teach us a profound lesson. Let $X = \{p, q, r\}$. Consider the topology $\mathcal{T} = \{\emptyset, \{r\}, \{p, q\}, X\}$. The closed sets are the complements: $\{\emptyset, \{r\}, \{p, q\}, X\}$. Notice that the singleton sets $\{p\}$ and $\{q\}$ are not closed. This means the space is not $T_1$. But is it regular? Let's check. Can we separate the point $p$ from the disjoint closed set $\{r\}$? Yes: put $p$ in the open set $\{p, q\}$ and $\{r\}$ in the open set $\{r\}$. They are disjoint. What about separating $r$ from the closed set $\{p, q\}$? Yes: put $r$ in $\{r\}$ and $\{p, q\}$ in $\{p, q\}$. They are disjoint. This space is regular! We've just constructed a space that is regular but not even $T_1$. This shatters the naive assumption of a simple linear hierarchy. The properties are more subtly related, which is why topologists are so careful with their definitions, often defining a ​​$T_3$ space​​ as one that is both regular and $T_1$.

The failure of these intuitive implications continues. Can we find a space where we can't separate two disjoint closed sets? Yes. Consider the real numbers with the ​​particular point topology​​ at $p=0$. A set is open if it's empty or if it contains 0. In this space, a set is closed if it's the whole line or if it doesn't contain 0. So, let's take two disjoint, non-empty closed sets, say $C_1 = (-\infty, 0)$ and $C_2 = (0, \infty)$. To separate them, we need an open set $U_1$ containing $C_1$ and an open set $U_2$ containing $C_2$. But for $U_1$ and $U_2$ to be open and non-empty, they both must contain 0. Thus, they can never be disjoint. This space is not normal.

The hierarchy of separation is a rich and complex tapestry, with each axiom capturing a different flavor of "niceness." The jump from $T_3$ (regular) to ​​$T_{3\frac{1}{2}}$​​ (completely regular or Tychonoff) introduces the idea of separating points from closed sets using continuous functions to $[0,1]$, a much stronger condition. And just like the jump from $T_1$ to regular, the implication is not reversible: there exist $T_3$ spaces that are not completely regular, adding another layer of texture to our understanding.

When we discover a "nice" property, like being Hausdorff or normal, a natural question arises: is this property inherited by smaller pieces of the space, or does it scale up when we build larger spaces? In other words, is the property contagious?

Let's start with subspaces. The Hausdorff property is wonderfully robust; any subspace of a Hausdorff space is also Hausdorff. If you can put bubbles around points in the big space, you can certainly do so in a smaller piece of it. But what about normality? Prepare for a shock. We can construct a space $X$ (the product of two special ordered sets) that is compact and Hausdorff, which guarantees it is normal. It's as well-behaved as you could wish. Now, we perform a tiny act of surgery: we remove a single point to get a subspace $Y$, known as the ​​Tychonoff plank​​. The result? The subspace $Y$ is no longer normal! There exist two disjoint sets within $Y$—think of them as the right-hand and top edges of a rectangle from which the top-right corner has been removed—that are closed in $Y$ but have become impossible to separate with disjoint open sets. Normality is a fragile, global property, not necessarily inherited by its parts.

What if we go the other way? If we start with a nice subspace, will its closure also be nice? Let $Y$ be a subspace and $\overline{Y}$ be its closure (the smallest closed set containing it). Let's build a space $X$ where the subspace $Y = \mathbb{N}$ has the discrete topology—every point is an open set, making it perfectly Hausdorff. For its closure, we add two new points, $a$ and $b$. We define the topology on $X$ such that any open set containing $a$ must also contain $b$. These two points are topologically inseparable, like two distinct locations that occupy the same "fuzzy" region of space. The closure $\overline{Y}$ is the entire space $X$, which is not Hausdorff. We started with a perfectly nice Hausdorff subspace, and its closure was not nice at all. Properties don't always spread outward.

Finally, let's consider building large spaces by multiplying smaller ones. A ​​second-countable​​ space is one that has a countable "address book," or basis, for its open sets. The real line $\mathbb{R}$ is second-countable (intervals with rational endpoints suffice). The plane $\mathbb{R}^2 = \mathbb{R} \times \mathbb{R}$ is also second-countable. In fact, any finite or even countably infinite product of second-countable spaces is second-countable. It seems like a very stable property. But this stability shatters when we cross the chasm from the countable to the uncountable. Consider the space $\{0, 1\}^{\mathbb{R}}$, an uncountable product of the simple two-point discrete space. The resulting space is so immense that no countable collection of basis sets could possibly describe all its open regions. It is not second-countable. This teaches us a fundamental lesson: the leap to uncountability changes everything.

One might intuitively think that having more open sets is always better. A topology that has more open sets is called ​​finer​​. Surely, a finer topology gives us more tools, more precision, and preserves the nice properties we had before?

This intuition is wrong. Consider the real line with its usual topology. It is ​​first-countable​​, meaning every point has a countable "local basis"—a countable collection of nested open sets that can approximate any neighborhood of that point. Now, let's make the topology finer by adding some very specific, "porous" open sets around the origin. We keep all the old open sets but also add new ones, like an open interval with a countable number of points removed. The new topology $\tau'$ is strictly finer than the old one $\tau$. But what have we done? At the origin, we have introduced so many new and bizarre open sets that no countable collection of neighborhoods can approximate all of them anymore. The space is no longer first-countable at the origin. By adding more open sets, we destroyed a desirable property. Having more open sets means a local basis has a much harder job to do.

This is the beauty of topology through the lens of counterexamples. Each strange space, each failed implication, is not a sign of chaos. It is a signpost, marking the true boundaries of a concept, teaching us to appreciate the precision of the definitions and the surprising, intricate, and deeply beautiful structure of the world of abstract space.

There is a wonderful story, perhaps apocryphal, about an engineer, a physicist, and a mathematician who are asked to build a fence around a flock of sheep using the least amount of material. The engineer, a practical soul, arranges the sheep in a tight circle and builds a fence around them. The physicist, seeking a more optimal solution, imagines an infinitely long fence and draws it taut around the flock, declaring that this encloses them with the minimum perimeter for a given area. The mathematician, however, simply builds a small fence around herself and declares, "I define myself to be on the outside."

In the world of topology, we are often the mathematicians in that story. We are not just concerned with the familiar shapes of our world, but with the very definition of "inside" and "outside," "near" and "far," "connected" and "separate." And just as the mathematician's clever definition reveals the power of abstraction, so too do we learn the most about our subject not when our intuition works, but when it spectacularly fails. The heroes of this story are the counterexamples.

These are not mere exceptions or unfortunate errors. They are the crucibles in which our theories are tested and refined. Like a bridge engineer who deliberately pushes a design to its breaking point to understand its true limits, we use counterexamples to discover the hidden assumptions and fragile boundaries of our mathematical ideas. They are the explorers' maps to the wild frontiers of logic, showing us where our familiar Euclidean intuition no longer applies. Let us take a journey through this gallery of beautiful "monsters," each of which tells a profound story about the fabric of space.

In our everyday experience, some ideas seem inextricably linked. If a place is "connected," meaning it's all one piece, it seems obvious that you can travel from any point within it to any other. But is this always true? Topology invites us to be more precise. "Connected" means you can't split the space into two disjoint, non-empty open sets. "Path-connected" means you can actually trace a continuous path, a function $\gamma$ from the interval $[0, 1]$, between any two points. Surely, these must be the same thing?

Let's look at a famous resident of the topological zoo: the ​​Topologist's Sine Curve​​. Imagine the graph of the function $y = \sin(1/x)$ for $x$ between $0$ and $1$. As $x$ gets closer to zero, the function oscillates with ever-increasing, frantic energy. Now, let's add the vertical line segment on the $y$-axis from $-1$ to $1$, which the curve gets infinitely close to. The entire shape, curve plus line segment, is connected. You cannot tear it into two separate open pieces. It is, in a very real sense, a single object.

But try to draw a path from a point on the wiggly curve, say at $x=1$, to a point on that vertical line segment, say the origin $(0,0)$. To do so, your path would have to follow the curve's oscillations. As it nears the $y$-axis, it would have to wiggle up and down faster and faster, traversing an infinite number of oscillations in a finite amount of time. Such a journey is impossible for any continuous path. The limit points on the $y$-axis are "reachable" in a topological sense (they are in the closure of the curve) but not in a practical, path-following sense. Here, in this simple picture, a vast canyon opens between the idea of being connected and being path-connected. The counterexample doesn't break mathematics; it illuminates it, forcing us to recognize these two concepts as distinct rungs on the ladder of topological structure.

Another intuitive link is between the "size" and "complexity" of a space. A space is separable if it has a countable "skeleton" of points that is dense, like the rational numbers $\mathbb{Q}$ within the real numbers $\mathbb{R}$. Any open set, no matter how small, must touch this skeleton. This seems to imply that the space can't be "too spacious." For instance, it must satisfy the countable chain condition (c.c.c.), meaning you can't cram an uncountable number of non-overlapping open "bubbles" into it. And indeed, it is a straightforward exercise to prove that every separable space is c.c.c..

But what about the other way around? If a space is c.c.c.—if it has no room for an uncountable collection of disjoint bubbles—must it be separable? Must it have a countable skeleton? For decades, this was a deep and open question. The answer, it turns out, is no. The counterexample is a far more elusive creature than the sine curve: the ​​Suslin line​​. Its construction is a marvel of mathematical logic, but its properties are what matter to us. It is a line-like object that is c.c.c. but stubbornly refuses to be separable. The existence of such a line, it turns out, is independent of the standard axioms of mathematics (ZFC); we can build consistent mathematical universes where it exists and others where it does not. This is a profound moment! A question that began in a topology course about the structure of space has led us to the very foundations of logic and set theory. The counterexample here is not just a clever shape; it is a gateway to understanding the limits of what we can prove.

Some of the most instructive counterexamples come from taking a familiar object, like the real number line, and giving its topology a subtle twist. Consider the ​​Sorgenfrey line​​, where our basic open sets are not the familiar open intervals $(a, b)$ but half-open intervals of the form $[a, b)$. This tiny change—including the left endpoint—unleashes a cascade of strange behaviors.

The Sorgenfrey line is a paragon of certain virtues. It is first-countable (every point has a countable collection of "shrinking" neighborhoods), and it is even completely normal (a very strong separation property). It is also a Baire space, meaning it is topologically "robust" and cannot be decomposed into a countable union of "thin" closed sets. With such a fine resume of properties, you would be forgiven for thinking it must be a metrizable space—a space where the topology could be generated by some notion of distance.

But it is not. The reason is a classic mismatch of properties. The Sorgenfrey line is separable (the rationals are still dense), but it is not second-countable (it does not have a countable basis for its entire topology). In any metric space, these two properties are equivalent. The fact that the Sorgenfrey line possesses one but not the other is a definitive proof that no "ruler" can be defined on it that gives back its $[a,b)$ topology. Our list of "nice" properties, however long, was simply not the right one to guarantee metrizability.

The story gets even stranger when we consider the product of two Sorgenfrey lines: the ​​Sorgenfrey plane​​. In mathematics, we often hope that "nice" properties are preserved when we build new objects from old ones. The Sorgenfrey line is completely normal, a property so nice it's inherited by all its subspaces. One might hope the Sorgenfrey plane, built from two such fine specimens, would be at least normal.

It is not. The plane contains a notorious "anti-diagonal" line consisting of points $(x, -x)$. Within this line, consider the points where $x$ is rational and the points where $x$ is irrational. These two sets, let's call them $A$ and $B$, are both closed in the Sorgenfrey plane's topology, and they are disjoint. In a normal space, you should be able to find two disjoint open "sleeves," one containing $A$ and the other containing $B$. But in the Sorgenfrey plane, this is impossible. The points of $A$ and $B$ are so intricately intermingled that any open set containing all of $A$ is doomed to touch any open set containing all of $B$.

This failure has immediate and dramatic consequences. The great ​​Tietze Extension Theorem​​ promises that in a normal space, any continuous real-valued function defined on a closed subset can be smoothly extended to the entire space. On the Sorgenfrey plane, we can define a simple continuous function on the closed set $A \cup B$: let $f(p) = 0$ for points in $A$ and $f(p) = 1$ for points in $B$. The failure to separate $A$ and $B$ is precisely the reason this function cannot be extended to a continuous function on the whole plane. The counterexample doesn't just say a theorem fails; it provides a beautiful, concrete reason why the theorem's hypotheses are essential.

Many concepts in geometry and physics are built on the idea of local-to-global principles. If a space is "nice" in every small neighborhood, we hope it will be "nice" globally. The ​​long line​​ is a startling counterexample to this hope. It is constructed by taking the first uncountable ordinal $\omega_1$ and inserting an open interval $(0,1)$ between each ordinal and its successor. The result is a space that, from the perspective of any single point, looks exactly like the real number line. It is a Hausdorff, linearly ordered space.

Yet, globally, it is monstrous. It is so "long" that it fails to be paracompact. Paracompactness is a somewhat technical but crucial property. It guarantees that any open cover of the space can be "shrunk" to a more manageable one that is locally finite—meaning any given point is only contained in a finite number of the refined open sets. This property is the bedrock on which partitions of unity are built, a tool that is absolutely essential for doing calculus on manifolds and is fundamental to fields like differential geometry and general relativity. The long line, despite being locally identical to the well-behaved real line, is so pathologically long that it cannot support these essential global structures. It teaches us that uncountability is a powerful force, and that local information is not always sufficient to tame it.

Our tour could go on. We could visit an uncountable set with the ​​discrete topology​​, a space so "separated" that every point is its own open set. It is developable (a property related to metrizability) but not separable, another broken implication. We could look at an uncountable set with the ​​cofinite topology​​, where only sets with finite complements are open. This space is separable and T1, but not second-countable, and certainly not Hausdorff, providing a basket of counterexamples to potential metrization theorems and challenging our intuition about connectedness. We could even examine a simple ​​two-point set with the trivial topology​​, which is "zero-dimensional" by one definition but remains stubbornly connected, showing that even our concept of dimension is a multi-faceted jewel.

What is the moral of this story? That topology is a minefield of exceptions? Not at all. The lesson is far more beautiful. These counterexamples are the guideposts that have forced us to create sharper tools, clearer definitions, and more powerful theorems. By understanding precisely why the Sorgenfrey plane is not normal, we understand with perfect clarity why normality is the key to the Tietze Extension Theorem. By seeing the chasm between connectedness and path-connectedness in the topologist's sine curve, we appreciate them as distinct, useful concepts.

The joy of science is not in being right, but in discovering the exact nature of how we are wrong. Every counterexample is a puzzle, a surprise, and a gift. They are the friction that polishes our understanding, and in their strange and unexpected forms, we see the true, deep, and often astonishing beauty of logical structure.