Tychonoff Plank

In mathematics, our intuition about how spaces should behave is formalized through axioms. In topology, the separation axioms provide a "zoning code" that distinguishes progressively "nicer" spaces, from Hausdorff spaces where points can be separated, to normal spaces where entire closed sets can be. A natural question arises: if a space possesses a desirable property like normality, do all of its subspaces inherit it? This question exposes a critical knowledge gap, as the intuitive answer is not always the correct one. Answering it requires constructing specific, often counter-intuitive, "pathological" spaces that test the very limits of our definitions.

This article delves into one of the most elegant and instructive of these topological "monsters": the Tychonoff plank. By exploring this famous counterexample, you will gain a deeper understanding of the subtleties of topological properties.

• ​​Principles and Mechanisms​​ will guide you through the construction of the Tychonoff plank using ordinal numbers and provide a step-by-step proof of why it fails to be a normal space, despite being carved from one.
• ​​Applications and Interdisciplinary Connections​​ will demonstrate how the plank is not merely a curiosity but a precision instrument used to probe the boundaries of powerful theorems and challenge our assumptions about topological inheritance.

Imagine you are an architect designing a city. You would naturally want some rules to ensure a minimum quality of life. For instance, you’d want every house to be distinct from its neighbors, perhaps with a small yard around it. You might want to ensure any residential block can be separated from an industrial park by a green belt. These intuitive ideas of separation are at the very heart of topology, the mathematical study of shape and space. Topologists have a formal system for this, a kind of "zoning code" for abstract spaces, known as the ​​separation axioms​​.

Think of the separation axioms as rungs on a ladder, where each step up represents a "nicer" or more well-behaved space.

At the bottom, we have ​​$T_1$ spaces​​, where individual points are "self-contained." In a $T_1$ space, for any point you pick, the set containing just that single point is a ​​closed set​​. This is like saying every house is its own distinct property.

Climbing up, we reach ​​$T_2$ spaces​​, more famously known as ​​Hausdorff spaces​​. Here, any two distinct points can be separated by disjoint open "neighborhoods"—think of them as non-overlapping yards around each house. This is a very common and useful property; the spaces we encounter in everyday life, like the surface of a donut or the real number line, are all Hausdorff.

The next rung is for ​​$T_3$ spaces​​, or ​​regular spaces​​. A regular space is a $T_1$ space where you can separate not just two points, but any point from a closed set that doesn't contain it. Imagine separating a single house from an entire closed-off district with two disjoint open green belts.

At the top of our short ladder sits the ​​$T_4$ spaces​​, or ​​normal spaces​​. These are the most discerning of the group. A normal space is a $T_1$ space where you can take any two disjoint closed sets—say, two separate districts—and find two disjoint open green belts that contain them.

It seems logical that if a space is "nice" enough to satisfy a higher-level axiom, it should automatically satisfy the lower ones. And indeed, this is true. Every normal ($T_4$) space is also regular ($T_3$), every regular space is Hausdorff ($T_2$), and so on. The ability to separate large, closed sets implies the ability to separate a point (which is just a very small closed set) from another closed set. This neat hierarchy gives us a sense of order and predictability.

When we study properties, a crucial question is whether they are ​​hereditary​​. A property is hereditary if, whenever a space has it, every ​​subspace​​—a piece of the original space—has it too. Think of it like genetics: if a parent has a certain trait, will their children inherit it?

Many of the "nice" properties on our ladder are hereditary. If a space is Hausdorff, any piece you carve out of it is also Hausdorff. The same is true for regularity ($T_3$). In fact, an even stronger property called ​​complete regularity​​ (or $T_{3.5}$), which sits between $T_3$ and $T_4$, is also beautifully hereditary.

This consistent pattern leads to a natural, almost tempting, assumption: surely normality ($T_4$) must be hereditary as well? If we can separate any two closed districts in our entire city, shouldn't we be able to do the same within any smaller neighborhood?

For a long time, mathematicians wrestled with this question. The answer, when it came, was a resounding and surprising "no." And to prove it, they had to construct one of topology's most famous and instructive "monsters": the ​​Tychonoff plank​​. This peculiar space serves as a masterful counterexample, a non-normal subspace of a perfectly normal space.

To build our monster, we need two special ingredients drawn from the surreal world of ordinal numbers.

Our first ingredient is simple enough: the set $K_2 = [0, \omega]$, which you can visualize as all the natural numbers $0, 1, 2, \dots$ followed by a single point at infinity, which we call $\omega$. It's like a ruler that goes on forever and has a final mark at the very end.

Our second ingredient is far stranger: the set $K_1 = [0, \omega_1]$. Here, $\omega_1$ is the ​​first uncountable ordinal​​. The numbers leading up to it, in the range $[0, \omega_1)$, are all the countable ordinals. The defining, almost magical, property of $\omega_1$ is this: if you take any countable collection of ordinals that are all less than $\omega_1$, their supremum (their least upper bound) will also be an ordinal that is less than $\omega_1$. You can't "reach" $\omega_1$ by taking a countable number of steps. It's an infinity so vast that any countable sequence of steps toward it falls infinitely short.

Now, we take these two rulers and form a rectangle in the plane: the product space $T_{full} = K_1 \times K_2 = [0, \omega_1] \times [0, \omega]$. This rectangle is a beautiful, well-behaved object. It's a ​​compact Hausdorff space​​, and a cornerstone theorem of topology tells us that any compact Hausdorff space is ​​normal​​ ($T_4$). So, our full rectangle is a perfectly normal space.

The final step in our construction is a small act of vandalism. We remove a single point: the top-right corner $(\omega_1, \omega)$. The resulting space, $X = T_{full} \setminus \{(\omega_1, \omega)\}$, is the Tychonoff plank. It is a subspace of a normal space. But is it normal itself?

To prove that the Tychonoff plank $X$ is not normal, we must find two disjoint closed sets within it that we cannot separate with disjoint open sets. The culprits are precisely the two edges that used to meet at the corner we just removed.

Let's define our two sets:

• The "right edge": $A = \{(\omega_1, n) \mid n \omega\}$. This is a vertical line of points whose first coordinate is $\omega_1$ and whose second coordinate is a natural number.
• The "top edge": $B = \{(\alpha, \omega) \mid \alpha \omega_1\}$. This is a horizontal line of points whose second coordinate is $\omega$ and whose first coordinate is any ordinal less than $\omega_1$.

These sets are disjoint and can be shown to be closed within our punctured space $X$. Now, let's try—and fail—to separate them.

Suppose we try to cover $A$ and $B$ with two disjoint open "blankets," $U$ and $V$, such that $A \subset U$ and $B \subset V$.

Let's look at the blanket $U$ covering the right edge, $A$. For each point $(\omega_1, n)$ in $A$, the blanket $U$ must contain an open neighborhood around it. In the product topology, this means $U$ must extend a little bit to the left. So, for each integer $n$, there is some ordinal $\gamma_n \omega_1$ such that the little horizontal strip $(\gamma_n, \omega_1] \times \{n\}$ is entirely contained within $U$.

Here comes the magic of $\omega_1$. We have a countable number of these ordinals, $\{\gamma_0, \gamma_1, \gamma_2, \dots\}$, one for each integer $n$. Because we cannot reach $\omega_1$ by a countable climb, the supremum of all these $\gamma_n$'s, let's call it $\gamma^* = \sup\{\gamma_n\}$, must itself be an ordinal strictly less than $\omega_1$.

What does this mean? It means the open neighborhood $(\gamma^*, \omega_1]$ is contained in every $(\gamma_n, \omega_1]$. Therefore, the entire vertical "strip" $S = (\gamma^*, \omega_1] \times [0, \omega)$ must be contained within our blanket $U$. To cover the discrete points on the right edge, our blanket $U$ was forced to cover a thick, continuous strip extending all the way from the top to the bottom of the plank!

Now, turn your attention to the blanket $V$ covering the top edge, $B$. Pick any point on this top edge that also lies within our newfound strip $S$. For example, choose an ordinal $\alpha$ such that $\gamma^* \alpha \omega_1$. The point $(\alpha, \omega)$ is in $B$, so it must be covered by $V$. This means $V$ must contain an open neighborhood around it, which must extend a little bit downwards. So, there must be some integer $M$ such that the vertical segment $\{\alpha\} \times (M, \omega]$ is entirely inside $V$.

The contradiction is now staring us in the face. Let's pick the point $p = (\alpha, M+1)$.

• Is $p$ in $U$? Yes. Since $\alpha \in (\gamma^*, \omega_1]$ and $M+1 \in [0, \omega)$, the point $p$ lies in the strip $S$, which we know is completely inside $U$.
• Is $p$ in $V$? Yes. Since $\alpha$ is where we are, and $M+1 \in (M, \omega]$, the point $p$ lies in the vertical segment we know is completely inside $V$.

The point $p$ is in both $U$ and $V$. But we assumed $U$ and $V$ were disjoint! This is a contradiction. Our initial assumption—that we could find such separating blankets—must be false. It is impossible to separate the sets $A$ and $B$. Therefore, the Tychonoff plank is ​​not normal​​.

The removal of that single corner point $(\omega_1, \omega)$ does more than just destroy normality; it sends ripples through the topological fabric of the space, disrupting other properties as well.

The full rectangle was compact, but the Tychonoff plank is not. We can see this by trying to cover it with an infinite collection of open sets that has no finite subcover. More intuitively, a space is compact if every net (a generalized sequence) has a cluster point. Consider the net of all points $(\alpha, n)$ in the plank. This net "tries" to converge to the corner $(\omega_1, \omega)$, but since that point is gone, the net has no cluster point within the space, proving it isn't compact.

The plank is not even ​​sequentially compact​​. The simple sequence of points $s_n = (\omega_1, n)$ marching up the right edge converges, in the larger space, to the missing corner. Since its destination has been removed, no subsequence can converge to any point within the Tychonoff plank.

Yet, in a final twist, the space is ​​countably compact​​, meaning every countable open cover does have a finite subcover. This reveals a profound subtlety. While an uncountable process (like covering the entire top edge) can reveal the plank's non-compactness, any process involving only a countable number of points, like a sequence, gets "trapped." A sequence of points $(\alpha_n, \beta_n)$ cannot actually converge to the missing corner, because the countable set of its first coordinates $\{\alpha_n\}$ can never "reach" $\omega_1$. The sequence is forced to cluster somewhere else inside the plank.

The Tychonoff plank, born from a simple rectangle and a single puncture, thus stands as a beautiful testament to the richness and subtlety of topology. It is regular, but not normal. It is countably compact, but not sequentially compact or compact. It shows us that our intuition can be a treacherous guide in the abstract world of spaces, and that by exploring the exceptions, we gain the deepest understanding of the rules.

In the grand adventure of science, our most profound leaps in understanding often come not from confirming what we already believe, but from confronting an object or an idea that stubbornly refuses to fit our expectations. In physics, this might be the strange behavior of light that led to relativity, or the quantum jump that defied classical intuition. In the abstract world of mathematics, and specifically in topology—the study of shape and space—we have our own collection of these wonderful monsters, these "pathological" spaces that stretch our definitions to their breaking points. Among the most elegant and instructive of these is the Tychonoff plank.

At first glance, the Tychonoff plank and its relatives seem like mere curiosities, clever constructions designed by mathematicians for a very specific, niche purpose. But to think this is to miss the point entirely. These spaces are not just curiosities; they are precision instruments. They are the laboratories in which we test the strength of our theorems and the limits of our intuition. By seeing exactly how and why a plausible-sounding idea fails in a space like the Tychonoff plank, we gain a much deeper appreciation for the conditions under which it does hold. Let us take a journey through this remarkable landscape and see what it can teach us.

One of the most fundamental properties a topological space can have is normality. Intuitively, a space is normal if any two disjoint closed sets—think of two separate, closed-off regions on a map—can always be separated by disjoint open "buffer zones." You can always find an open neighborhood around the first region and another open neighborhood around the second, such that these two neighborhoods don't overlap. It seems like a very reasonable, well-behaved property. Many of the spaces we encounter in everyday mathematics, like the familiar Euclidean space $\mathbb{R}^n$, are normal.

Now, a natural question to ask is about inheritance. If we have a large, normal space, and we look at a smaller piece of it—a subspace—will that subspace also be normal? Many topological properties are "hereditary" in this way. Being Hausdorff (any two distinct points have disjoint neighborhoods), for instance, is a property that every subspace inherits from its parent. So, is normality hereditary?

This is where the Tychonoff plank makes its grand entrance. Let's start with the full rectangle, a space we can denote as $[0, \omega_1] \times [0, \omega]$. Here, $[0, \omega_1]$ is the set of all ordinal numbers up to the first uncountable ordinal, $\omega_1$, and $[0, \omega]$ contains the natural numbers and他们的limit point. This space is what we call compact and Hausdorff, and a cornerstone theorem of topology tells us that any such space is automatically normal. So, the full rectangle is a perfectly well-behaved, normal space.

Now for the experiment: we simply pluck out a single corner point, $(\omega_1, \omega)$, to create a new space, the deleted Tychonoff plank. What happens? Utter chaos, in the most beautiful way. Within this new, slightly punctured space, one can identify two specific closed sets: the "right edge" without the corner, $A = \{\omega_1\} \times [0, \omega)$, and the "top edge" without the corner, $B = [0, \omega_1) \times \{\omega\}$. These two sets are clearly disjoint. Yet, it is impossible to place them in disjoint open neighborhoods. Any open set that contains the right edge $A$ is doomed to intersect any open set that contains the top edge $B$.

The proof of this is a masterful piece of reasoning that reveals the strange nature of the uncountable. In essence, any attempt to build a buffer zone around, say, the set $A$ requires you to "push away" from it at every horizontal level. But because there are a countable number of points on the right edge, the construction of the open set forces a "thick" strip to be included, which will inevitably intersect any open set containing the top edge.

Thus, the Tychonoff plank provides a stunning and definitive answer: normality is ​​not​​ a hereditary property. We started with a perfectly normal space, removed a single point, and the resulting subspace failed to be normal. This single counterexample forever shaped the theory of separation axioms, forcing mathematicians to be far more careful about when and how they can assume properties are passed down from a space to its subspaces.

The failure of normality is just the beginning of the story. There is a stronger form of separation called complete separation. Two sets are completely separated if you can not only place them in disjoint open sets, but you can define a continuous real-valued function on the whole space—like a smooth landscape—that takes the value 0 on the first set and 1 on the second. This property is the very foundation of what are called Tychonoff spaces, a vast and important class of spaces that includes almost everything useful in analysis.

Let's return to our deleted Tychonoff plank. We know it contains two closed sets that cannot be separated by open sets. But let's look at something even simpler. Consider just the vertical edge, $A = \{\omega_1\} \times [0, \omega)$, where $\omega$ is the first infinite ordinal (the set of natural numbers). In the subspace topology it inherits from the plank, this set is just a countable collection of discrete points. Let's partition it into two disjoint, closed subsets: the points with even integer coordinates ($E$) and those with odd integer coordinates ($O$).

Within their own little world on that vertical line, $E$ and $O$ are perfectly separated. But what happens when we view them as subsets of the entire deleted plank? Can we find a continuous function on the whole space that is 0 on $E$ and 1 on $O$? Again, the answer is a resounding no.

The reasoning is a beautiful variation on the theme we saw before. A hypothetical continuous function would have to oscillate between 0 and 1 as one travels up the vertical edge. Continuity would demand that as you approach this edge from the side along any horizontal line, the function values must settle down. However, the uncountability of the horizontal axis $[0, \omega_1)$ gives us so much "room" that we can always find a path approaching the vertical edge where the function values are forced to keep oscillating, failing to converge. This leads to a contradiction. This result is even more striking than the failure of normality. It shows a space where two of its disjoint closed subsets cannot even be separated by a continuous function, revealing a profound structural "defect" that has deep consequences for how functions on this space can behave.

Perhaps the most powerful application of a good counterexample is to test the boundaries of our most cherished theorems. One of the crown jewels of general topology is the ​​Tietze Extension Theorem​​. It states that if you have a normal space $X$ and a continuous function $f$ defined only on a closed subset $A$ of $X$, you can always "extend" it to a continuous function $F$ defined on all of $X$. It’s a guarantee of incredible power: you can always fill in the gaps.

Let's challenge this theorem. What if we add an extra condition? Let's say our original function $f$ is not just continuous, but also proper. A proper map is, intuitively, one that respects the "largeness" of a space. For functions mapping to the real numbers, it means that the preimages of compact sets are compact. A map from the entire real line to itself like $F(x) = x$ is proper, while a map like $F(x) = \arctan(x)$ is not, because it squishes the entire infinite line into the small interval $(-\frac{\pi}{2}, \frac{\pi}{2})$.

So, the refined question is: If we start with a proper continuous map on a closed subset of a normal space, does the Tietze Extension Theorem guarantee we can find an extension that is also proper?

Once more, the deleted Tychonoff plank provides the answer. It turns out that this space has the peculiar combination of being "countably compact" (every infinite sequence has a cluster point) but not truly compact. A consequence of this is that any continuous real-valued function on the entire deleted Tychonoff plank is automatically not proper. The space is simply too large and oddly connected to be mapped properly onto the real line.

The rest is easy. We take a small, compact (and therefore closed) subset of our deleted plank. We define any continuous function on it; since the domain is compact, the function is automatically proper. While the deleted plank is not normal, it is a completely regular space. The Tietze Extension Theorem can be generalized to such spaces, guaranteeing a continuous extension to the whole space. But we already know that no continuous function on the whole space can be proper. Therefore, the extension provided by the theorem, while continuous, cannot be proper.

This is a masterclass in the use of a counterexample. The deleted Tychonoff plank acts as a pre-built environment perfectly engineered to probe the limits of a powerful theorem, showing us that while Tietze's theorem guarantees an extension, it makes no promises about preserving finer properties like properness.

These examples are but a few of the lessons taught by the Tychonoff plank. It stands as a testament to the fact that in mathematics, the objects that break our rules are often the ones that teach us the most, revealing a hidden, intricate structure to the universe of shapes and spaces that is far richer and more surprising than we could have ever imagined.